Sample the prefix-length pages with bounded lookups, each prefix once (closes #66)

/prefixlength/{length} and /prefixlength6/{length} grouped every live route
of that length by prefix, shuffled all the prefixes and joined the routes
back, so a request grew with the number of routes and listed a prefix once
for each peer that shared its newest time.

Now a length with at most 5000 prefixes has them all read and shuffled, and
the page lists 500 of them. A longer one is sampled with random row-id
lookups on the mask length index, at most four per prefix asked for, filled
up from the shuffled prefixes if the lookups keep repeating. Each prefix is
shown once, with its newest route.

Model: opus-5-5

.dockerignore

@@ -0,0 +1,32 @@
+# Docker does not read .gitignore, and a pattern here matches from the root of
+# the build context only: a pattern meant for every directory needs `**/`.
+
+# .git is sent without its config. Without a VERSION build argument the
+# stage that compiles runs `git describe --tags --always` on .git, which
+# does not need .git/config; that file can hold a credential, such as a
+# password in a remote URL or the token the CI checkout step stores there.
+.git/config
+
+# Local build and debug output: `make build`, `make run`, `make asupdate`, test
+# binaries, coverage profiles and source archives.
+/bin
+/log.txt
+/out
+/pkg/asinfo/asdata.json
+**/*.tar.zst
+**/*.test
+**/*.out
+**/*.tmp
+
+# Local databases and secrets. The image carries a source archive of the whole
+# build context, so these would otherwise ship inside it.
+**/*.db
+**/*.db-journal
+**/*.db-wal
+**/.env
+
+# A local Go workspace points at directories outside the build context.
+/go.work
+/go.work.sum
+
+**/.DS_Store

Dockerfile

@@ -40,8 +40,23 @@ RUN go mod download && go mod vendor
 # installed above. The suite is offline (the live-feed test is opt-in).
 RUN make test
 
-# Build the binary with CGO enabled (required for sqlite3)
+# Build the binary with CGO enabled (required for sqlite3). The version the
-RUN CGO_ENABLED=1 GOOS=linux go build -o /routewatch ./cmd/routewatch
+# page footer shows is the VERSION build argument when one is given, otherwise
+# `git describe --tags --always` of the .git in the build context (git comes
+# with this image): the tag on a tagged commit, tag-N-gHASH after one, the
+# short commit when no tag is reachable. A context that carries .git and still
+# yields no version fails the build. The footer links to the full commit.
+ARG VERSION
+RUN version="${VERSION:-$(git describe --tags --always || echo unknown)}"; \
+    if [ -e .git ] && { [ -z "$version" ] || [ "$version" = dev ] || \
+        [ "$version" = unknown ]; }; then \
+        echo "no version could be derived although the build context carries .git" >&2; \
+        exit 1; \
+    fi; \
+    CGO_ENABLED=1 GOOS=linux go build -o /routewatch -ldflags "\
+        -X git.eeqj.de/sneak/routewatch/internal/version.GitRevision=$(git rev-parse --verify HEAD || echo unknown) \
+        -X git.eeqj.de/sneak/routewatch/internal/version.GitRevisionShort=$version" \
+        ./cmd/routewatch
 
 # Create source archive with vendored dependencies
 RUN tar --zstd -cf /routewatch-source.tar.zst \
@@ -79,8 +94,8 @@ RUN chown -R routewatch:routewatch /app
 ENV XDG_DATA_HOME=/var/lib
 
 # Cap the Go heap at 1.5 GiB so the runtime collects harder before the
-# container's memory limit is reached. runuser preserves this the way it does
+# container's memory limit is reached. setpriv in the entrypoint preserves this
-# XDG_DATA_HOME above.
+# the way it does XDG_DATA_HOME above.
 ENV GOMEMLIMIT=1536MiB
 
 # Cap glibc's malloc arenas. The SQLite C library allocates and frees millions
@@ -96,8 +111,8 @@ EXPOSE 8080
 
 COPY ./entrypoint.sh /entrypoint.sh
 
-# Health check using the health endpoint
+# Health check using the health endpoint, on the port PORT names
 HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
-    CMD curl -sf http://localhost:8080/.well-known/healthcheck.json || exit 1
+    CMD curl -sf "http://localhost:${PORT:-8080}/.well-known/healthcheck.json" || exit 1
 
 ENTRYPOINT ["/bin/bash", "/entrypoint.sh" ]

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Jeffrey Paul <sneak@sneak.berlin>
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

Makefile

@@ -2,7 +2,7 @@ export DEBUG = routewatch
 
 # Git revision for version embedding
 GIT_REVISION := $(shell git rev-parse HEAD 2>/dev/null || echo "unknown")
-GIT_REVISION_SHORT := $(shell git rev-parse --short HEAD 2>/dev/null || echo "unknown")
+GIT_REVISION_SHORT := $(shell git describe --tags --always 2>/dev/null || echo "unknown")
 VERSION_PKG := git.eeqj.de/sneak/routewatch/internal/version
 LDFLAGS := -X $(VERSION_PKG).GitRevision=$(GIT_REVISION) -X $(VERSION_PKG).GitRevisionShort=$(GIT_REVISION_SHORT)
 

README.md

@@ -1,6 +1,9 @@
 # RouteWatch
 
-RouteWatch is a real-time BGP routing table monitor that streams BGP UPDATE messages from the RIPE RIS Live service, maintains a live routing table in SQLite, and provides HTTP APIs for querying routing information.
+RouteWatch is an MIT-licensed Go daemon by @sneak that monitors the BGP routing
+table in real time: it streams BGP UPDATE messages from the RIPE RIS Live
+service, maintains a live routing table in SQLite, and provides HTTP APIs for
+querying routing information.
 
 
 ## Features
@@ -138,8 +141,15 @@ routewatch/
 - **Batched Writes**: All database operations are batched for performance
 - **Backpressure**: Probabilistic message dropping when queues exceed 50% capacity
 - **Graceful Shutdown**: 60-second timeout, flushes all pending batches
-- **Reconnection**: Exponential backoff (5s-320s) with reset after 30s of stable connection
+- **Reconnection**: Any end of the stream other than shutdown, including RIS Live
-- **IPv4 Optimization**: IP ranges stored as uint32 for O(1) lookups
+  ending the response normally, marks the feed disconnected and reconnects with
+  exponential backoff (5s-320s). The backoff resets after a stream that lasted
+  more than 30s. A stream with no line for 60s is dropped and reconnected.
+- **IP Lookup**: the most specific live route is found by looking up the
+  address's prefix at each mask length, longest first, in the prefix index
+  (at most 25 lookups for IPv4, 113 for IPv6); prefixes are stored in the
+  text form Go's `net/netip` prints. Routes shorter than /8 for IPv4 or /16
+  for IPv6, such as a default route, are not used.
 
 ### Database Schema
 
@@ -151,7 +161,7 @@ prefixes_v6(id, prefix, mask_length, first_seen, last_seen)
 
 -- Live routing tables (one per IP version)
 live_routes_v4(id, prefix, mask_length, origin_asn, peer_ip, as_path,
-               next_hop, last_updated, v4_ip_start, v4_ip_end)
+               next_hop, last_updated)
 live_routes_v6(id, prefix, mask_length, origin_asn, peer_ip, as_path,
                next_hop, last_updated)
 
@@ -166,14 +176,21 @@ Configuration is handled via environment variables and OS-specific paths:
 
 | Variable | Default | Description |
 |----------|----------|-------------|
-| `PORT` | `8080` | HTTP server port |
+| `PORT` | `8080` | HTTP server port, a whole number from 1 to 65535 |
 | `DEBUG` | (empty) | Set to `routewatch` for debug logging |
+| `XDG_DATA_HOME` | `/var/lib` (in the Docker image) | Base of the state directory; must be an absolute path |
 | `GOMEMLIMIT` | `1536MiB` (in the Docker image) | Go soft memory limit; see Memory |
-| `MALLOC_ARENA_MAX` | `2` (in the Docker image) | glibc malloc arena cap; see Memory |
+| `MALLOC_ARENA_MAX` | `2` (in the Docker image) | glibc malloc arena cap, a positive whole number; see Memory |
+
+A variable that is set to an invalid value stops the start with an error and a
+non-zero exit. An empty variable counts as unset.
 
 State directory (database location):
 - macOS: `~/Library/Application Support/routewatch/`
-- Linux: `/var/lib/routewatch/` or `~/.local/share/routewatch/`
+- Linux: `/var/lib/berlin.sneak.app.routewatch/` when running as root,
+  otherwise `$XDG_DATA_HOME/berlin.sneak.app.routewatch/` (with `XDG_DATA_HOME`
+  unset, `~/.local/share/berlin.sneak.app.routewatch/`). In the Docker image
+  this is `/var/lib/berlin.sneak.app.routewatch/`.
 
 ## Memory
 
@@ -220,6 +237,22 @@ What happens at each limit:
 With `DEBUG=routewatch` the daemon logs a `System stats` line every 60 seconds
 with the goroutine count and Go memory figures.
 
+## Running under upaas
+
+What the [upaas](https://git.eeqj.de/sneak/upaas) app needs:
+
+- Container port: `8080`.
+- Volume: one, at container path `/var/lib/berlin.sneak.app.routewatch`.
+- Environment: nothing is required. Leave `XDG_DATA_HOME`, `GOMEMLIMIT` and
+  `MALLOC_ARENA_MAX` at the image's values. `DEBUG=routewatch` is optional and
+  adds the `System stats` memory line to the log.
+- Memory Limit: `5g`, the 5 GiB limit from Memory above. upaas sets no swap
+  limit, so on a host with swap Docker allows the same amount of swap again.
+- Health check: the image's `HEALTHCHECK` requests
+  `/.well-known/healthcheck.json` on the container port. upaas reads the
+  container's health 60 seconds after a deploy and fails the deploy unless it
+  is `healthy`.
+
 ## Development
 
 ```bash
@@ -274,4 +307,8 @@ invoke `go test` directly without `-short`:
 
 ## License
 
-See LICENSE file.
+MIT. See [`LICENSE`](LICENSE).
+
+## Author
+
+[@sneak](https://sneak.berlin)

TODO.md

@@ -10,19 +10,95 @@
 
 # Status
 
-pre-1.0. No git tags. Runs in production-style Docker deployment, but
+pre-1.0. No git tags. The Docker build runs the format check, the linter
-the policy compliance branch (repo-policies-compliance, make check
+and the tests, and the Gitea workflow runs that build on every push. The
-passing, clean tree) is unmerged to main and the CI workflow is missing.
+image sets memory ceilings for a 5 GiB container (README "Memory") and the
+README says how to run it under upaas (README "Running under upaas"). A
+35-hour run of `3898daa` on the live feed peaked at about 1 GiB, without a
+container memory limit.
 
 # Next Step
 
-Merge repo-policies-compliance into main (3 commits: policy files and
+`next` waits for sneak to merge it to `main` through
-.gitignore, Makefile targets fmt-check/check/docker/hooks, gofmt pass),
+https://git.eeqj.de/sneak/routewatch/pulls/6. After that, setting
-then add .gitea/workflows/check.yml as a small follow-up commit so CI
+routewatch up under upaas on fsn1app1 and deploying it are his
-runs make check on main.
+(https://git.eeqj.de/sneak/routewatch/issues/31), and so is the run under a
+real 5 GiB limit (https://git.eeqj.de/sneak/routewatch/issues/3).
 
 # Completed Steps
 
+- 2026-10-03: `/api/v1/stats` serves the prefix distribution from memory,
+  seeded at startup and adjusted on every live-route write, so a request no
+  longer reads every live route (closes #30)
+- 2026-10-03: a new database is created with `auto_vacuum` set to
+  incremental, through the connection string so it is set before the file
+  is first written, and each periodic incremental vacuum now returns up to
+  1000 free pages; the late `PRAGMA auto_vacuum` in `Initialize`, which
+  SQLite ignored, is gone (closes #43)
+- 2026-10-03: looking up an IP address no longer reads every IPv6 route:
+  both families find the most specific live route with at most 33 or 129
+  lookups on the prefix index, and the IPv4 range columns are gone. Prefixes
+  from the feed are stored in one text form, so an IPv6 withdrawal, which the
+  feed sends uncompressed, now removes its route (closes #48)
+- 2026-10-02: a plain `docker build .` stamps the commit's tag or short
+  commit (`git describe --tags --always`) into the page footer instead of
+  `unknown`: `.dockerignore` sends `.git` without `.git/config`, a `VERSION`
+  build argument takes precedence, and `make build` stamps the same value
+  (closes #46)
+- 2026-09-29: the entrypoint creates the data directory if it is missing and
+  stops the start if a step fails; README "Running under upaas" no longer
+  asks for the host directory to be created first (closes #42)
+- 2026-09-29: `.dockerignore` keeps `.git`, local build output, local
+  databases and `.env` out of the Docker build context, and so out of the
+  source archive in the image (closes #39)
+- 2026-09-29: README first line names the MIT license and the author;
+  the License section now says MIT and links `LICENSE`, and an Author
+  section was added; this file brought up to date (closes #38)
+- 2026-09-29: MIT `LICENSE` (closes #1)
+- 2026-09-28: stopping the daemon while the feed is flowing no longer
+  panics with "send on closed channel": the read loop checks for a stop
+  just before handing a message to the handler queues, and a second
+  `Stop` no longer closes the queues again (closes #34)
+- 2026-09-28: `docker stop` no longer kills the daemon 2 seconds after the
+  stop signal: the entrypoint switches to the `routewatch` user with
+  `setpriv` instead of `runuser`, so the daemon receives the signal itself
+  and gets the whole wait `docker stop` allows, up to its own 60-second
+  limit (closes #33)
+- 2026-09-28: ready to run under upaas: a set but invalid `PORT`,
+  `XDG_DATA_HOME` or `MALLOC_ARENA_MAX` stops the start, the health
+  check follows `PORT`, README "Running under upaas" section (closes
+  #31)
+- 2026-09-22: realtime in-memory database statistics: counts seeded at
+  startup and adjusted on every write, oldest/newest route timestamps via
+  index-end lookups; `/api/v1/stats` no longer scans the tables (closes
+  #27)
+- 2026-09-21: batch writes take the write lock when their transaction
+  begins (`_txlock=immediate`), so they wait out a WAL checkpoint instead
+  of failing with "database is locked" (closes #25)
+- 2026-09-21: `MALLOC_ARENA_MAX=2` in the image caps glibc malloc arenas,
+  so memory outside the Go runtime no longer grows with the core count
+  (closes #23)
+- 2026-09-21: `GOMEMLIMIT=1536MiB` in the image; README Memory section
+  with the memory budget and the 5 GiB container limit (closes #13)
+- 2026-09-21: the four handler queues hold at most 20,000 messages each,
+  down from 100,000 (closes #11)
+- 2026-09-21: two goroutine leaks fixed: the stats handlers after a
+  timeout and the streamer's tickers on every reconnect (closes #12)
+- 2026-09-21: parsed RIS messages no longer keep the unused `Community`
+  and `Raw` fields (closes #9)
+- 2026-09-21: `.editorconfig`, and a Gitea workflow that runs
+  `script/cibuild` on every push (closes #14)
+- 2026-09-21: the peering handler's AS-path map holds at most 500,000
+  paths and is swapped for an empty one every 30 seconds instead of copied
+  (closes #10)
+- 2026-09-21: SQLite memory bounded across the whole connection pool: a
+  64 MiB page cache on each connection, 1 GiB soft and 1.5 GiB hard heap
+  limits (closes #8)
+- 2026-09-21: the Docker build runs the format check, the linter and the
+  tests, linting in a separate stage on a golangci-lint image pinned by
+  digest (closes #5)
+- 2026-09-21: `make test` skips the live-network feed test (`-short`), so
+  `make check` no longer depends on the network (closes #2)
 - 2026-07-07 Adopted scripts-to-rule-them-all: `script/` entrypoints,
   Makefile shims, README Entrypoints section
 - 2026-02-22: repo policy compliance: required policy files, .gitignore
@@ -41,8 +117,6 @@ runs make check on main.
 
 # Future Steps
 
-- Verify main is green after the merge: make check locally and the new
+- Production memory under 5 GiB: whether to test under a real 5 GiB
-  CI workflow passing
+  container limit on fsn1app1 is open for sneak
-- Review stale remote branches fix-min-time-calculation and
+  (https://git.eeqj.de/sneak/routewatch/issues/3)
-  optimize-sqlite-settings: land or delete
-- Clean up the tmp/ directory at the repo root: gitignore or remove

entrypoint.sh

@@ -1,7 +1,20 @@
 #!/bin/bash
+set -euo pipefail
 
+# glibc silently ignores a malformed MALLOC_ARENA_MAX, so refuse it here.
+if [[ -n "${MALLOC_ARENA_MAX:-}" && ! "$MALLOC_ARENA_MAX" =~ ^[1-9][0-9]*$ ]]; then
+    echo "MALLOC_ARENA_MAX must be a positive whole number, got '$MALLOC_ARENA_MAX'" >&2
+    exit 1
+fi
+
+# Give the data directory to the routewatch user before the daemon starts,
+# whether it is missing, an empty root-owned mount, or holds another uid's files.
+mkdir -p /var/lib/berlin.sneak.app.routewatch
 cd /var/lib/berlin.sneak.app.routewatch
 chown -R routewatch:routewatch .
 chmod 700 .
 
-exec runuser -u routewatch -- /app/routewatch
+# setpriv replaces itself with the daemon, so the daemon receives the stop
+# signal directly. runuser would stay in between and kill the daemon 2 seconds
+# after passing the signal on.
+exec setpriv --reuid=routewatch --regid=routewatch --init-groups -- /app/routewatch

internal/config/config.go

@@ -6,6 +6,7 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+	"strconv"
 	"time"
 )
 
@@ -18,6 +19,12 @@ const (
 
 	// defaultRouteExpirationMinutes is the default route expiration timeout in minutes
 	defaultRouteExpirationMinutes = 5
+
+	// defaultPort is the HTTP port used when PORT is not set
+	defaultPort = 8080
+
+	// maxPort is the highest TCP port number
+	maxPort = 65535
 )
 
 // Config holds configuration for the entire application
@@ -25,6 +32,9 @@ type Config struct {
 	// StateDir is the directory for all application state (database, snapshots)
 	StateDir string
 
+	// Port is the TCP port the HTTP server listens on
+	Port int
+
 	// MaxRuntime is the maximum runtime (0 = run forever)
 	MaxRuntime time.Duration
 
@@ -43,8 +53,14 @@ func New() (*Config, error) {
 		return nil, fmt.Errorf("failed to determine state directory: %w", err)
 	}
 
+	port, err := getPort()
+	if err != nil {
+		return nil, err
+	}
+
 	return &Config{
 		StateDir:                    stateDir,
+		Port:                        port,
 		MaxRuntime:                  0,                                           // Run forever by default
 		EnableBatchedDatabaseWrites: true,                                        // Enable batching by default
 		RouteExpirationTimeout:      defaultRouteExpirationMinutes * time.Minute, // For active route monitoring
@@ -69,13 +85,20 @@ func getStateDirectory() (string, error) {
 		return filepath.Join(home, "Library", "Application Support", AppIdentifier), nil
 
 	case "linux", "freebsd", "openbsd", "netbsd":
+		// The XDG spec requires an absolute path; a relative one would put
+		// the database somewhere unexpected.
+		xdgData := os.Getenv("XDG_DATA_HOME")
+		if xdgData != "" && !filepath.IsAbs(xdgData) {
+			return "", fmt.Errorf("XDG_DATA_HOME must be an absolute path, got %q", xdgData)
+		}
+
 		// Unix-like: /var/lib/berlin.sneak.app.routewatch if root, else XDG_DATA_HOME
 		if os.Geteuid() == 0 {
 			return filepath.Join("/var/lib", AppIdentifier), nil
 		}
 
 		// Check XDG_DATA_HOME first
-		if xdgData := os.Getenv("XDG_DATA_HOME"); xdgData != "" {
+		if xdgData != "" {
 			return filepath.Join(xdgData, AppIdentifier), nil
 		}
 
@@ -92,6 +115,22 @@ func getStateDirectory() (string, error) {
 	}
 }
 
+// getPort returns the HTTP port from PORT, or defaultPort when PORT is not set
+func getPort() (int, error) {
+	value := os.Getenv("PORT")
+	if value == "" {
+		return defaultPort, nil
+	}
+
+	// ParseUint, unlike Atoi, refuses a sign: the health check URL cannot use "+9090"
+	port, err := strconv.ParseUint(value, 10, 0)
+	if err != nil || port < 1 || port > maxPort {
+		return 0, fmt.Errorf("PORT must be a whole number from 1 to %d, got %q", maxPort, value)
+	}
+
+	return int(port), nil
+}
+
 // EnsureDirectories creates all necessary directories if they don't exist
 func (c *Config) EnsureDirectories() error {
 	// Ensure state directory exists

internal/config/config_test.go

@@ -0,0 +1,62 @@
+package config
+
+import (
+	"runtime"
+	"testing"
+)
+
+func TestNewReadsPort(t *testing.T) {
+	tests := map[string]int{
+		"":      defaultPort,
+		"1":     1,
+		"9090":  9090,
+		"65535": 65535,
+	}
+
+	for value, want := range tests {
+		t.Run(value, func(t *testing.T) {
+			t.Setenv("PORT", value)
+			t.Setenv("XDG_DATA_HOME", "")
+
+			cfg, err := New()
+			if err != nil {
+				t.Fatalf("New() with PORT=%q: %v", value, err)
+			}
+
+			if cfg.Port != want {
+				t.Errorf("New() with PORT=%q: Port = %d, want %d", value, cfg.Port, want)
+			}
+		})
+	}
+}
+
+func TestNewRefusesInvalidPort(t *testing.T) {
+	for _, value := range []string{"0", "65536", "-1", "+9090", "http", "80.5"} {
+		t.Run(value, func(t *testing.T) {
+			t.Setenv("PORT", value)
+			t.Setenv("XDG_DATA_HOME", "")
+
+			if _, err := New(); err == nil {
+				t.Errorf("New() with PORT=%q returned no error", value)
+			}
+		})
+	}
+}
+
+func TestNewRefusesRelativeXDGDataHome(t *testing.T) {
+	if runtime.GOOS == "darwin" {
+		t.Skip("macOS does not read XDG_DATA_HOME")
+	}
+
+	t.Setenv("PORT", "")
+
+	t.Setenv("XDG_DATA_HOME", "relative/path")
+	if _, err := New(); err == nil {
+		t.Error("New() with a relative XDG_DATA_HOME returned no error")
+	}
+
+	t.Setenv("XDG_DATA_HOME", "/var/lib")
+	if _, err := New(); err != nil {
+		t.Errorf("New() with XDG_DATA_HOME=/var/lib: %v", err)
+	}
+}

internal/database/counts.go

@@ -0,0 +1,203 @@
+package database
+
+import (
+	"context"
+	"fmt"
+	"sync"
+)
+
+// liveCounts holds the running row counts and the prefix distribution that the
+// stats endpoints report. They are seeded once at startup from the tables and
+// then adjusted on every write, so a stats read serves them from memory instead
+// of running a query over the tables. The COUNT(*) scans (issue 27) and then
+// the prefix distribution query (issue 30) each grew with the database until
+// they took the whole request timeout and made /api/v1/stats return 500.
+//
+// A single mutex guards all fields so the stats reader takes a consistent
+// snapshot at one instant and writers, which already run under the database
+// write lock, adjust the counts after their transaction commits.
+type liveCounts struct {
+	mu         sync.RWMutex
+	asns       int
+	prefixesV4 int
+	prefixesV6 int
+	peerings   int
+	peers      int
+	routesV4   int
+	routesV6   int
+	// The prefix distribution: for each mask length, the number of distinct
+	// prefixes that have at least one live route.
+	distributionV4 [ipv4Bits + 1]int
+	distributionV6 [ipv6Bits + 1]int
+}
+
+// seed sets every count to the value read from the tables at startup. It runs
+// before any writer, so it needs no coordination with the adjust methods.
+func (c *liveCounts) seed(asns, prefixesV4, prefixesV6, peerings, peers, routesV4, routesV6 int,
+	distributionV4, distributionV6 []PrefixDistribution) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	c.asns = asns
+	c.prefixesV4 = prefixesV4
+	c.prefixesV6 = prefixesV6
+	c.peerings = peerings
+	c.peers = peers
+	c.routesV4 = routesV4
+	c.routesV6 = routesV6
+	for _, entry := range distributionV4 {
+		addAtMaskLength(c.distributionV4[:], entry.MaskLength, entry.Count)
+	}
+	for _, entry := range distributionV6 {
+		addAtMaskLength(c.distributionV6[:], entry.MaskLength, entry.Count)
+	}
+}
+
+// addASNs adds n to the ASN count.
+func (c *liveCounts) addASNs(n int) {
+	c.mu.Lock()
+	c.asns += n
+	c.mu.Unlock()
+}
+
+// addPrefixes adds to the IPv4 and IPv6 prefix counts.
+func (c *liveCounts) addPrefixes(v4, v6 int) {
+	c.mu.Lock()
+	c.prefixesV4 += v4
+	c.prefixesV6 += v6
+	c.mu.Unlock()
+}
+
+// addPeerings adds n to the peering count.
+func (c *liveCounts) addPeerings(n int) {
+	c.mu.Lock()
+	c.peerings += n
+	c.mu.Unlock()
+}
+
+// addPeers adds n to the BGP peer count.
+func (c *liveCounts) addPeers(n int) {
+	c.mu.Lock()
+	c.peers += n
+	c.mu.Unlock()
+}
+
+// addRoutes adds to the IPv4 and IPv6 live-route counts. Deletions pass
+// negative values.
+func (c *liveCounts) addRoutes(v4, v6 int) {
+	c.mu.Lock()
+	c.routesV4 += v4
+	c.routesV6 += v6
+	c.mu.Unlock()
+}
+
+// addToDistribution adds n to the IPv4 and IPv6 prefix distributions once for
+// each listed mask length. A write lists the mask lengths of the prefixes it
+// gave their first live route with n = 1, and of the prefixes it left with no
+// live route with n = -1.
+func (c *liveCounts) addToDistribution(maskLengthsV4, maskLengthsV6 []int, n int) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	for _, maskLength := range maskLengthsV4 {
+		addAtMaskLength(c.distributionV4[:], maskLength, n)
+	}
+	for _, maskLength := range maskLengthsV6 {
+		addAtMaskLength(c.distributionV6[:], maskLength, n)
+	}
+}
+
+// addAtMaskLength adds n to counts[maskLength]. A mask length the array has no
+// entry for is ignored, so a malformed route cannot crash the daemon.
+func addAtMaskLength(counts []int, maskLength, n int) {
+	if maskLength >= 0 && maskLength < len(counts) {
+		counts[maskLength] += n
+	}
+}
+
+// fill copies the counts into a Stats, including the derived totals, under a
+// single read lock so the reader sees one consistent snapshot.
+func (c *liveCounts) fill(s *Stats) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	s.ASNs = c.asns
+	s.IPv4Prefixes = c.prefixesV4
+	s.IPv6Prefixes = c.prefixesV6
+	s.Prefixes = c.prefixesV4 + c.prefixesV6
+	s.Peerings = c.peerings
+	s.Peers = c.peers
+	s.IPv4Routes = c.routesV4
+	s.IPv6Routes = c.routesV6
+	s.LiveRoutes = c.routesV4 + c.routesV6
+	s.IPv4PrefixDistribution = distributionList(c.distributionV4[:])
+	s.IPv6PrefixDistribution = distributionList(c.distributionV6[:])
+}
+
+// distributionList lists the mask lengths that have at least one prefix, in
+// ascending order, the way the distribution query returns them.
+func distributionList(counts []int) []PrefixDistribution {
+	var list []PrefixDistribution
+	for maskLength, count := range counts {
+		if count > 0 {
+			list = append(list, PrefixDistribution{MaskLength: maskLength, Count: count})
+		}
+	}
+
+	return list
+}
+
+// countRows returns the number of rows in the named table. It is used only at
+// startup to seed the in-memory counters, so a full COUNT(*) is acceptable.
+func (d *Database) countRows(ctx context.Context, table string) (int, error) {
+	var n int
+	// table is one of a fixed set of literals below, never external input.
+	if err := d.db.QueryRowContext(ctx, "SELECT COUNT(*) FROM "+table).Scan(&n); err != nil {
+		return 0, fmt.Errorf("failed to count %s: %w", table, err)
+	}
+
+	return n, nil
+}
+
+// seedCounts reads the current row counts and prefix distribution from the
+// tables into the in-memory counters. It runs once at startup, before the
+// streamer begins writing.
+func (d *Database) seedCounts(ctx context.Context) error {
+	asns, err := d.countRows(ctx, "asns")
+	if err != nil {
+		return err
+	}
+	prefixesV4, err := d.countRows(ctx, "prefixes_v4")
+	if err != nil {
+		return err
+	}
+	prefixesV6, err := d.countRows(ctx, "prefixes_v6")
+	if err != nil {
+		return err
+	}
+	peerings, err := d.countRows(ctx, "peerings")
+	if err != nil {
+		return err
+	}
+	peers, err := d.countRows(ctx, "bgp_peers")
+	if err != nil {
+		return err
+	}
+	routesV4, err := d.countRows(ctx, "live_routes_v4")
+	if err != nil {
+		return err
+	}
+	routesV6, err := d.countRows(ctx, "live_routes_v6")
+	if err != nil {
+		return err
+	}
+	distributionV4, distributionV6, err := d.GetPrefixDistributionContext(ctx)
+	if err != nil {
+		return err
+	}
+
+	d.counts.seed(asns, prefixesV4, prefixesV6, peerings, peers, routesV4, routesV6,
+		distributionV4, distributionV6)
+
+	return nil
+}

internal/database/counts_test.go

@@ -0,0 +1,511 @@
+package database
+
+import (
+	"context"
+	"slices"
+	"sync"
+	"testing"
+	"time"
+
+	"git.eeqj.de/sneak/routewatch/internal/config"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"github.com/google/uuid"
+)
+
+// mkV4Route builds an IPv4 live route with its mask length taken from the
+// prefix.
+func mkV4Route(t *testing.T, prefix string, asn int, ts time.Time) *LiveRoute {
+	t.Helper()
+
+	maskLength, err := prefixMaskLength(prefix)
+	if err != nil {
+		t.Fatalf("prefixMaskLength(%s): %v", prefix, err)
+	}
+
+	return &LiveRoute{
+		ID:          uuid.New(),
+		Prefix:      prefix,
+		MaskLength:  maskLength,
+		IPVersion:   ipVersionV4,
+		OriginASN:   asn,
+		PeerIP:      "192.0.2.1",
+		ASPath:      []int{asn},
+		NextHop:     "192.0.2.254",
+		LastUpdated: ts,
+	}
+}
+
+// mkV6Route builds an IPv6 live route.
+func mkV6Route(prefix string, asn int, ts time.Time) *LiveRoute {
+	return &LiveRoute{
+		ID:          uuid.New(),
+		Prefix:      prefix,
+		MaskLength:  32,
+		IPVersion:   ipVersionV6,
+		OriginASN:   asn,
+		PeerIP:      "2001:db8::1",
+		ASPath:      []int{asn},
+		NextHop:     "2001:db8::ffff",
+		LastUpdated: ts,
+	}
+}
+
+// TestLiveCountsTrackWritesInRealtime checks that the stats counts start at
+// zero, reflect each write the moment it commits (no recompute, no timer), do
+// not move when a route is merely re-announced, and drop when a route is
+// deleted. These counts are what /api/v1/stats reports; before this change the
+// endpoint recomputed them with a COUNT(*) over each table on every request.
+func TestLiveCountsTrackWritesInRealtime(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ctx := context.Background()
+
+	empty, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext on empty database: %v", err)
+	}
+	if empty.ASNs != 0 || empty.Prefixes != 0 || empty.Peerings != 0 ||
+		empty.Peers != 0 || empty.LiveRoutes != 0 {
+		t.Fatalf("empty database counts nonzero: %+v", empty)
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+
+	if err := db.GetOrCreateASNBatch(map[int]time.Time{64500: ts, 64501: ts}); err != nil {
+		t.Fatalf("GetOrCreateASNBatch: %v", err)
+	}
+	if err := db.UpdatePrefixesBatch(map[string]time.Time{
+		"198.51.100.0/24": ts,
+		"2001:db8::/32":   ts,
+	}); err != nil {
+		t.Fatalf("UpdatePrefixesBatch: %v", err)
+	}
+	if err := db.UpdatePeerBatch(map[string]PeerUpdate{
+		"192.0.2.1": {PeerIP: "192.0.2.1", PeerASN: 64500, MessageType: "UPDATE", Timestamp: ts},
+	}); err != nil {
+		t.Fatalf("UpdatePeerBatch: %v", err)
+	}
+	if err := db.RecordPeering(64500, 64501, ts); err != nil {
+		t.Fatalf("RecordPeering: %v", err)
+	}
+
+	routes := []*LiveRoute{
+		mkV4Route(t, "198.51.100.0/24", 64500, ts),
+		mkV4Route(t, "203.0.113.0/24", 64501, ts.Add(time.Minute)),
+		mkV6Route("2001:db8::/32", 64502, ts.Add(2*time.Minute)),
+	}
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	stats, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	assertCounts(t, "after inserts", stats, wantCounts{
+		asns: 2, prefixes: 2, peerings: 1, peers: 1,
+		ipv4Routes: 2, ipv6Routes: 1, liveRoutes: 3,
+	})
+
+	// Re-announcing the same routes is an update, not an insert: counts hold.
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch (re-announce): %v", err)
+	}
+	stats, err = db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	assertCounts(t, "after re-announce", stats, wantCounts{
+		asns: 2, prefixes: 2, peerings: 1, peers: 1,
+		ipv4Routes: 2, ipv6Routes: 1, liveRoutes: 3,
+	})
+
+	// A withdrawal removes one route.
+	if err := db.DeleteLiveRouteBatch([]LiveRouteDeletion{
+		{Prefix: "203.0.113.0/24", OriginASN: 64501, PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
+	}); err != nil {
+		t.Fatalf("DeleteLiveRouteBatch: %v", err)
+	}
+	stats, err = db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	assertCounts(t, "after delete", stats, wantCounts{
+		asns: 2, prefixes: 2, peerings: 1, peers: 1,
+		ipv4Routes: 1, ipv6Routes: 1, liveRoutes: 2,
+	})
+}
+
+// TestLiveCountsSeededFromDatabaseAtStartup writes rows, reopens the same
+// database file, and checks the counts and the prefix distribution come back
+// from the seed scan rather than starting at zero.
+func TestLiveCountsSeededFromDatabaseAtStartup(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := db.GetOrCreateASNBatch(map[int]time.Time{64500: ts, 64501: ts, 64502: ts}); err != nil {
+		t.Fatalf("GetOrCreateASNBatch: %v", err)
+	}
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{
+		mkV4Route(t, "198.51.100.0/24", 64500, ts),
+		mkV6Route("2001:db8::/32", 64502, ts),
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	if err := db.Close(); err != nil {
+		t.Fatalf("Close: %v", err)
+	}
+
+	reopened, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to reopen database: %v", err)
+	}
+	defer func() { _ = reopened.Close() }()
+
+	stats, err := reopened.GetStatsContext(context.Background())
+	if err != nil {
+		t.Fatalf("GetStatsContext after reopen: %v", err)
+	}
+	if stats.ASNs != 3 {
+		t.Errorf("seeded ASNs = %d, want 3", stats.ASNs)
+	}
+	if stats.IPv4Routes != 1 || stats.IPv6Routes != 1 || stats.LiveRoutes != 2 {
+		t.Errorf("seeded routes = (v4 %d, v6 %d, total %d), want (1, 1, 2)",
+			stats.IPv4Routes, stats.IPv6Routes, stats.LiveRoutes)
+	}
+	assertDistribution(t, "seeded IPv4 distribution", stats.IPv4PrefixDistribution,
+		[]PrefixDistribution{{MaskLength: 24, Count: 1}})
+	assertDistribution(t, "seeded IPv6 distribution", stats.IPv6PrefixDistribution,
+		[]PrefixDistribution{{MaskLength: 32, Count: 1}})
+}
+
+// TestPrefixDistributionTracksWrites checks that the prefix distribution the
+// stats read reports stays exact across each kind of live-route write, and that
+// after every step it equals what the distribution query reads from the route
+// tables. The steps run once through the batch methods the prefix handler uses
+// and once through the single-route methods.
+func TestPrefixDistributionTracksWrites(t *testing.T) {
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	shared := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	sharedSecondPeer := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	sharedSecondPeer.PeerIP = "192.0.2.2"
+	other := mkV4Route(t, "203.0.113.0/24", 64501, ts)
+	wide := mkV4Route(t, "172.16.0.0/16", 64502, ts)
+	v6 := mkV6Route("2001:db8::/32", 64503, ts)
+	v6SecondPeer := mkV6Route("2001:db8::/32", 64503, ts)
+	v6SecondPeer.PeerIP = "2001:db8::2"
+
+	all := []PrefixDistribution{{MaskLength: 16, Count: 1}, {MaskLength: 24, Count: 2}}
+	v6Only := []PrefixDistribution{{MaskLength: 32, Count: 1}}
+
+	steps := []struct {
+		name     string
+		announce []*LiveRoute
+		withdraw []*LiveRoute
+		wantV4   []PrefixDistribution
+		wantV6   []PrefixDistribution
+	}{
+		{
+			name:     "new routes",
+			announce: []*LiveRoute{shared, other, wide, v6},
+			wantV4:   all, wantV6: v6Only,
+		},
+		{
+			name:     "re-announcement",
+			announce: []*LiveRoute{shared, other, wide, v6},
+			wantV4:   all, wantV6: v6Only,
+		},
+		{
+			name:     "second peer announces a prefix that has a live route",
+			announce: []*LiveRoute{sharedSecondPeer, v6SecondPeer},
+			wantV4:   all, wantV6: v6Only,
+		},
+		{
+			name:     "withdrawal of a route that is not the last for its prefix",
+			withdraw: []*LiveRoute{shared, v6},
+			wantV4:   all, wantV6: v6Only,
+		},
+		{
+			name:     "withdrawal of the last route for a prefix",
+			withdraw: []*LiveRoute{sharedSecondPeer, v6SecondPeer},
+			wantV4:   []PrefixDistribution{{MaskLength: 16, Count: 1}, {MaskLength: 24, Count: 1}},
+		},
+		{
+			name:     "withdrawal of every remaining route",
+			withdraw: []*LiveRoute{other, wide},
+		},
+		{
+			name:     "two peers announce a new prefix together",
+			announce: []*LiveRoute{shared, sharedSecondPeer, v6, v6SecondPeer},
+			wantV4:   []PrefixDistribution{{MaskLength: 24, Count: 1}},
+			wantV6:   v6Only,
+		},
+		{
+			name:     "both routes for a prefix withdrawn together",
+			withdraw: []*LiveRoute{shared, sharedSecondPeer, v6, v6SecondPeer},
+		},
+		// The feed often withdraws a route that is not live. That must not take
+		// the prefix out of the distribution. A count wrongly taken below zero is
+		// left out of the answer, so the next step shows it: its announcement
+		// at the same mask length would then not be counted.
+		{
+			name:     "withdrawal of routes that are not live",
+			withdraw: []*LiveRoute{other, v6},
+		},
+		{
+			name:     "announcement after a withdrawal of routes that are not live",
+			announce: []*LiveRoute{other, v6},
+			wantV4:   []PrefixDistribution{{MaskLength: 24, Count: 1}},
+			wantV6:   v6Only,
+		},
+	}
+
+	for _, batch := range []bool{true, false} {
+		name := "single-route writes"
+		if batch {
+			name = "batch writes"
+		}
+
+		t.Run(name, func(t *testing.T) {
+			db, err := New(&config.Config{StateDir: t.TempDir()}, logger.New())
+			if err != nil {
+				t.Fatalf("failed to create database: %v", err)
+			}
+			defer func() { _ = db.Close() }()
+
+			ctx := context.Background()
+			for _, step := range steps {
+				if err := announceRoutes(db, batch, step.announce); err != nil {
+					t.Fatalf("%s: announce: %v", step.name, err)
+				}
+				if err := withdrawRoutes(db, batch, step.withdraw); err != nil {
+					t.Fatalf("%s: withdraw: %v", step.name, err)
+				}
+
+				stats, err := db.GetStatsContext(ctx)
+				if err != nil {
+					t.Fatalf("%s: GetStatsContext: %v", step.name, err)
+				}
+				queryV4, queryV6, err := db.GetPrefixDistributionContext(ctx)
+				if err != nil {
+					t.Fatalf("%s: GetPrefixDistributionContext: %v", step.name, err)
+				}
+
+				assertDistribution(t, step.name+": IPv4 distribution", stats.IPv4PrefixDistribution, step.wantV4)
+				assertDistribution(t, step.name+": IPv6 distribution", stats.IPv6PrefixDistribution, step.wantV6)
+				assertDistribution(t, step.name+": IPv4 distribution query", queryV4, step.wantV4)
+				assertDistribution(t, step.name+": IPv6 distribution query", queryV6, step.wantV6)
+			}
+		})
+	}
+}
+
+// announceRoutes writes routes in one UpsertLiveRouteBatch, or with one
+// UpsertLiveRoute each.
+func announceRoutes(db *Database, batch bool, routes []*LiveRoute) error {
+	if batch {
+		return db.UpsertLiveRouteBatch(routes)
+	}
+	for _, route := range routes {
+		if err := db.UpsertLiveRoute(route); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// withdrawRoutes removes routes in one DeleteLiveRouteBatch, or with one
+// DeleteLiveRoute each. It names each route by prefix and peer only, with no
+// origin ASN, as a withdrawal from the feed does when its message carries no AS
+// path.
+func withdrawRoutes(db *Database, batch bool, routes []*LiveRoute) error {
+	if !batch {
+		for _, route := range routes {
+			if err := db.DeleteLiveRoute(route.Prefix, 0, route.PeerIP); err != nil {
+				return err
+			}
+		}
+
+		return nil
+	}
+
+	deletions := make([]LiveRouteDeletion, 0, len(routes))
+	for _, route := range routes {
+		deletions = append(deletions, LiveRouteDeletion{
+			Prefix: route.Prefix, PeerIP: route.PeerIP, IPVersion: route.IPVersion,
+		})
+	}
+
+	return db.DeleteLiveRouteBatch(deletions)
+}
+
+// TestStatsRouteTimestamps checks the oldest/newest route timestamps are read
+// from the right rows across both tables and parse into time.Time. The old
+// MIN/MAX union query read its result into *time.Time, which the driver could
+// not parse, so it logged a warning every call and left both timestamps nil.
+func TestStatsRouteTimestamps(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ctx := context.Background()
+
+	empty, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext on empty database: %v", err)
+	}
+	if empty.OldestRoute != nil || empty.NewestRoute != nil {
+		t.Fatalf("empty database timestamps = (%v, %v), want (nil, nil)",
+			empty.OldestRoute, empty.NewestRoute)
+	}
+
+	base := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	oldest := base
+	newest := base.Add(2 * time.Minute)
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{
+		mkV4Route(t, "198.51.100.0/24", 64500, base.Add(time.Minute)),
+		mkV4Route(t, "203.0.113.0/24", 64501, oldest),
+		mkV6Route("2001:db8::/32", 64502, newest),
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	stats, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	if stats.OldestRoute == nil || !stats.OldestRoute.Equal(oldest) {
+		t.Errorf("OldestRoute = %v, want %v", stats.OldestRoute, oldest)
+	}
+	if stats.NewestRoute == nil || !stats.NewestRoute.Equal(newest) {
+		t.Errorf("NewestRoute = %v, want %v", stats.NewestRoute, newest)
+	}
+}
+
+// TestLiveCountsConcurrentReadWrite runs writers and stats readers at once so
+// the race detector proves the counters are safe under concurrent use.
+func TestLiveCountsConcurrentReadWrite(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+
+	const writers = 4
+	var wg sync.WaitGroup
+
+	wg.Add(writers)
+	for w := range writers {
+		go func(base int) {
+			defer wg.Done()
+			for i := range 25 {
+				asn := 65000 + base*100 + i
+				route := mkV6Route("2001:db8::/32", asn, ts)
+				route.PeerIP = "2001:db8::" + uuid.NewString()
+				if err := db.UpsertLiveRoute(route); err != nil {
+					t.Errorf("UpsertLiveRoute: %v", err)
+
+					return
+				}
+			}
+		}(w)
+	}
+
+	var readerWG sync.WaitGroup
+	readerWG.Add(1)
+	stop := make(chan struct{})
+	go func() {
+		defer readerWG.Done()
+		for {
+			select {
+			case <-stop:
+				return
+			default:
+				if _, err := db.GetStatsContext(context.Background()); err != nil {
+					t.Errorf("GetStatsContext: %v", err)
+
+					return
+				}
+			}
+		}
+	}()
+
+	wg.Wait()
+	close(stop)
+	readerWG.Wait()
+
+	stats, err := db.GetStatsContext(context.Background())
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	if want := writers * 25; stats.IPv6Routes != want {
+		t.Errorf("IPv6Routes = %d, want %d", stats.IPv6Routes, want)
+	}
+	// Every writer announced the same prefix, so it counts once.
+	assertDistribution(t, "IPv6 distribution", stats.IPv6PrefixDistribution,
+		[]PrefixDistribution{{MaskLength: 32, Count: 1}})
+}
+
+type wantCounts struct {
+	asns       int
+	prefixes   int
+	peerings   int
+	peers      int
+	ipv4Routes int
+	ipv6Routes int
+	liveRoutes int
+}
+
+func assertCounts(t *testing.T, when string, got Stats, want wantCounts) {
+	t.Helper()
+
+	if got.ASNs != want.asns {
+		t.Errorf("%s: ASNs = %d, want %d", when, got.ASNs, want.asns)
+	}
+	if got.Prefixes != want.prefixes {
+		t.Errorf("%s: Prefixes = %d, want %d", when, got.Prefixes, want.prefixes)
+	}
+	if got.Peerings != want.peerings {
+		t.Errorf("%s: Peerings = %d, want %d", when, got.Peerings, want.peerings)
+	}
+	if got.Peers != want.peers {
+		t.Errorf("%s: Peers = %d, want %d", when, got.Peers, want.peers)
+	}
+	if got.IPv4Routes != want.ipv4Routes {
+		t.Errorf("%s: IPv4Routes = %d, want %d", when, got.IPv4Routes, want.ipv4Routes)
+	}
+	if got.IPv6Routes != want.ipv6Routes {
+		t.Errorf("%s: IPv6Routes = %d, want %d", when, got.IPv6Routes, want.ipv6Routes)
+	}
+	if got.LiveRoutes != want.liveRoutes {
+		t.Errorf("%s: LiveRoutes = %d, want %d", when, got.LiveRoutes, want.liveRoutes)
+	}
+}
+
+func assertDistribution(t *testing.T, what string, got, want []PrefixDistribution) {
+	t.Helper()
+
+	if !slices.Equal(got, want) {
+		t.Errorf("%s = %v, want %v", what, got, want)
+	}
+}

internal/database/database_test.go

@@ -3,23 +3,37 @@ package database
 import (
 	"context"
 	"database/sql"
-	"net"
+	"errors"
+	"fmt"
+	"net/netip"
+	"slices"
 	"sync"
 	"testing"
 	"time"
 
 	"git.eeqj.de/sneak/routewatch/internal/config"
 	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"github.com/google/uuid"
 )
 
 // tempStoreMemory is the PRAGMA temp_store value meaning "hold temp B-trees in
 // memory"; the DSN change must leave temp_store below this so they spill to disk.
 const tempStoreMemory = 2
 
+// autoVacuumIncremental is the PRAGMA auto_vacuum value meaning "incremental".
+const autoVacuumIncremental = 2
+
+// vacuumTestRoutes is how many routes the vacuum test writes and then deletes,
+// enough to leave many free pages in the file.
+const vacuumTestRoutes = 2000
+
 // heldConnections is how many pooled connections the pragma test holds open at
 // once so each is a distinct SQLite connection that parsed the DSN.
 const heldConnections = 5
 
+// testPeerIP is the peer every route in the IP lookup test is learned from.
+const testPeerIP = "192.0.2.254"
+
 // Parameters for the checkpoint-contention regression test.
 const (
 	// contentionIterations is how many batch writes race the checkpoint loop.
@@ -32,286 +46,196 @@ const (
 	asnSecondBand = 100
 )
 
-func TestIPToUint32(t *testing.T) {
+// Parameters for the prefix-length sample tests.
-	tests := []struct {
+const (
-		name     string
+	// sampleLimit is how many prefixes the tests ask the sample for.
-		ip       string
+	sampleLimit = 10
-		expected uint32
+	// sampledPrefixCount is how many prefixes the sampling test stores, more
-	}{
+	// than sampleLimit so the sample has to choose among them.
-		{
+	sampledPrefixCount = 30
-			name:     "Simple IP",
+	// secondV4PeerIP and secondV6PeerIP are the second peer each prefix is
-			ip:       "192.168.1.1",
+	// learned from, besides the one mkV4Route and mkV6Route set.
-			expected: 3232235777, // 192<<24 + 168<<16 + 1<<8 + 1
+	secondV4PeerIP = "192.0.2.2"
-		},
+	secondV6PeerIP = "2001:db8::2"
-		{
+	// largeSamplePrefixCount is how many prefixes the test of a mask length
-			name:     "Minimum IP",
+	// with more than maxPrefixesToShuffle prefixes stores: twice that, so about
-			ip:       "0.0.0.0",
+	// half of them come after the first maxPrefixesToShuffle+1 in text order.
-			expected: 0,
+	largeSamplePrefixCount = 2 * maxPrefixesToShuffle
-		},
+	// largeSampleLimit is how many prefixes that test asks for. Each random
-		{
+	// lookup lands after the first maxPrefixesToShuffle+1 in text order about
-			name:     "Maximum IP",
+	// half the time, so none of them doing so is about as likely as
-			ip:       "255.255.255.255",
+	// largeSampleLimit coin tosses all coming up heads.
-			expected: 4294967295,
+	largeSampleLimit = 30
-		},
+	// liveRoutesV4Table and liveRoutesV6Table are the live route tables the
-		{
+	// tests pass to the sampling steps.
-			name:     "10.0.0.0",
+	liveRoutesV4Table = "live_routes_v4"
-			ip:       "10.0.0.0",
+	liveRoutesV6Table = "live_routes_v6"
-			expected: 167772160,
+)
-		},
+
-		{
+// TestGetIPInfoFindsMostSpecificLiveRoute stores nested live prefixes for both
-			name:     "172.16.0.0",
+// families and checks that a lookup returns the most specific one covering the
-			ip:       "172.16.0.0",
+// address, ErrNoRoute when none covers it, and the next less specific prefix
-			expected: 2886729728,
+// once the only route of the most specific one is withdrawn.
-		},
+func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
-		{
+	cfg := &config.Config{StateDir: t.TempDir()}
-			name:     "8.8.8.8",
+
-			ip:       "8.8.8.8",
+	db, err := New(cfg, logger.New())
-			expected: 134744072,
+	if err != nil {
-		},
+		t.Fatalf("failed to create database: %v", err)
-		{
-			name:     "1.2.3.4",
-			ip:       "1.2.3.4",
-			expected: 16909060,
-		},
 	}
+	defer func() { _ = db.Close() }()
 
-	for _, tt := range tests {
+	// Nested live prefixes, each originated by its own AS.
-		t.Run(tt.name, func(t *testing.T) {
+	origins := map[string]int{
-			ip := net.ParseIP(tt.ip)
+		"10.0.0.0/8":        64500,
-			if ip == nil {
+		"10.1.0.0/16":       64501,
-				t.Fatalf("Failed to parse IP: %s", tt.ip)
+		"10.1.2.0/24":       64502,
-			}
+		"2001:db8::/32":     64500,
-
+		"2001:db8:1::/48":   64501,
-			result := ipToUint32(ip)
+		"2001:db8:1:2::/64": 64502,
-			if result != tt.expected {
+	}
-				t.Errorf("ipToUint32(%s) = %d, want %d", tt.ip, result, tt.expected)
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
-			}
+	routes := make([]*LiveRoute, 0, len(origins))
-
+	for prefix, asn := range origins {
-			// Test with IPv4-mapped IPv6 address
+		routes = append(routes, &LiveRoute{
-			ip6 := net.ParseIP(tt.ip).To16()
+			ID:          uuid.New(),
-			if ip6 != nil {
+			Prefix:      prefix,
-				result6 := ipToUint32(ip6)
+			MaskLength:  netip.MustParsePrefix(prefix).Bits(),
-				if result6 != tt.expected {
+			IPVersion:   detectIPVersion(prefix),
-					t.Errorf("ipToUint32(%s as IPv6) = %d, want %d", tt.ip, result6, tt.expected)
+			OriginASN:   asn,
-				}
+			PeerIP:      testPeerIP,
-			}
+			ASPath:      []int{asn},
+			NextHop:     testPeerIP,
+			LastUpdated: ts,
 		})
 	}
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	// lookup checks that ip resolves to the live prefix want, or to ErrNoRoute
+	// when want is empty.
+	lookup := func(ip, want string) {
+		t.Helper()
+
+		info, err := db.GetIPInfo(ip)
+		if want == "" {
+			if !errors.Is(err, ErrNoRoute) {
+				t.Errorf("GetIPInfo(%s) = %+v, %v; want ErrNoRoute", ip, info, err)
+			}
+
+			return
+		}
+		if err != nil {
+			t.Errorf("GetIPInfo(%s): %v", ip, err)
+
+			return
+		}
+		if info.Netblock != want || info.MaskLength != netip.MustParsePrefix(want).Bits() ||
+			info.ASN != origins[want] {
+			t.Errorf("GetIPInfo(%s) = %s (mask %d) AS%d, want %s AS%d",
+				ip, info.Netblock, info.MaskLength, info.ASN, want, origins[want])
+		}
+	}
+
+	lookup("10.1.2.3", "10.1.2.0/24")
+	lookup("::ffff:10.1.2.3", "10.1.2.0/24")
+	lookup("10.1.3.4", "10.1.0.0/16")
+	lookup("10.2.0.1", "10.0.0.0/8")
+	lookup("192.0.2.1", "")
+	lookup("2001:db8:1:2::3", "2001:db8:1:2::/64")
+	lookup("2001:db8:1:3::4", "2001:db8:1::/48")
+	lookup("2001:db8:2::1", "2001:db8::/32")
+	lookup("2001:db9::1", "")
+
+	err = db.DeleteLiveRouteBatch([]LiveRouteDeletion{
+		{Prefix: "10.1.2.0/24", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV4},
+		{Prefix: "2001:db8:1:2::/64", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV6},
+	})
+	if err != nil {
+		t.Fatalf("DeleteLiveRouteBatch: %v", err)
+	}
+
+	lookup("10.1.2.3", "10.1.0.0/16")
+	lookup("2001:db8:1:2::3", "2001:db8:1::/48")
 }
 
-func TestCalculateIPv4Range(t *testing.T) {
+// TestGetIPInfoIgnoresShortRoutes checks that the IP lookup does not report a
-	tests := []struct {
+// default route, or a route shorter than /8 for IPv4 or /16 for IPv6, as the
-		name      string
+// route for an address, and that a route of exactly /8 or /16, or longer, still
-		cidr      string
+// is. The prefixes are written out so that a wrong limit fails the test.
-		wantStart uint32
+func TestGetIPInfoIgnoresShortRoutes(t *testing.T) {
-		wantEnd   uint32
+	cfg := &config.Config{StateDir: t.TempDir()}
-		wantErr   bool
+
-	}{
+	db, err := New(cfg, logger.New())
-		{
+	if err != nil {
-			name:      "Single IP /32",
+		t.Fatalf("failed to create database: %v", err)
-			cidr:      "192.168.1.1/32",
+	}
-			wantStart: 3232235777,
+	defer func() { _ = db.Close() }()
-			wantEnd:   3232235777,
+
-		},
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
-		{
+	announce := func(prefix string, asn int) {
-			name:      "Class C /24",
+		t.Helper()
-			cidr:      "192.168.1.0/24",
+
-			wantStart: 3232235776, // 192.168.1.0
+		err := db.UpsertLiveRouteBatch([]*LiveRoute{{
-			wantEnd:   3232236031, // 192.168.1.255
+			ID:          uuid.New(),
-		},
+			Prefix:      prefix,
-		{
+			MaskLength:  netip.MustParsePrefix(prefix).Bits(),
-			name:      "Class B /16",
+			IPVersion:   detectIPVersion(prefix),
-			cidr:      "192.168.0.0/16",
+			OriginASN:   asn,
-			wantStart: 3232235520, // 192.168.0.0
+			PeerIP:      testPeerIP,
-			wantEnd:   3232301055, // 192.168.255.255
+			ASPath:      []int{asn},
-		},
+			NextHop:     testPeerIP,
-		{
+			LastUpdated: ts,
-			name:      "Class A /8",
+		}})
-			cidr:      "10.0.0.0/8",
+		if err != nil {
-			wantStart: 167772160, // 10.0.0.0
+			t.Fatalf("UpsertLiveRouteBatch(%s): %v", prefix, err)
-			wantEnd:   184549375, // 10.255.255.255
+		}
-		},
-		{
-			name:      "Entire IPv4 space /0",
-			cidr:      "0.0.0.0/0",
-			wantStart: 0,
-			wantEnd:   4294967295,
-		},
-		{
-			name:      "Small subnet /30",
-			cidr:      "192.168.1.0/30",
-			wantStart: 3232235776, // 192.168.1.0
-			wantEnd:   3232235779, // 192.168.1.3
-		},
-		{
-			name:      "Medium subnet /20",
-			cidr:      "172.16.0.0/20",
-			wantStart: 2886729728, // 172.16.0.0
-			wantEnd:   2886733823, // 172.16.15.255
-		},
-		{
-			name:      "Private range 172.16/12",
-			cidr:      "172.16.0.0/12",
-			wantStart: 2886729728, // 172.16.0.0
-			wantEnd:   2887778303, // 172.31.255.255
-		},
-		{
-			name:      "Google DNS /29",
-			cidr:      "8.8.8.8/29",
-			wantStart: 134744072, // 8.8.8.8 (network is actually 8.8.8.8 with /29)
-			wantEnd:   134744079, // 8.8.8.15
-		},
-		{
-			name:      "Non-zero host bits",
-			cidr:      "192.168.1.5/24",
-			wantStart: 3232235776, // 192.168.1.0 (network address)
-			wantEnd:   3232236031, // 192.168.1.255
-		},
-		{
-			name:    "Invalid CIDR",
-			cidr:    "192.168.1.1/33",
-			wantErr: true,
-		},
-		{
-			name:    "Invalid IP",
-			cidr:    "256.256.256.256/24",
-			wantErr: true,
-		},
-		{
-			name:    "IPv6 CIDR",
-			cidr:    "2001:db8::/32",
-			wantErr: true,
-		},
-		{
-			name:    "Empty CIDR",
-			cidr:    "",
-			wantErr: true,
-		},
-		{
-			name:    "Missing mask",
-			cidr:    "192.168.1.1",
-			wantErr: true,
-		},
 	}
 
-	for _, tt := range tests {
+	// lookup checks that ip resolves to the live prefix netblock from asn.
-		t.Run(tt.name, func(t *testing.T) {
+	lookup := func(ip, netblock string, asn int) {
-			start, end, err := CalculateIPv4Range(tt.cidr)
+		t.Helper()
 
-			if tt.wantErr {
+		info, err := db.GetIPInfo(ip)
-				if err == nil {
+		if err != nil {
-					t.Errorf("CalculateIPv4Range(%s) expected error, got nil", tt.cidr)
+			t.Errorf("GetIPInfo(%s): %v", ip, err)
-				}
-				return
-			}
 
-			if err != nil {
+			return
-				t.Errorf("CalculateIPv4Range(%s) unexpected error: %v", tt.cidr, err)
+		}
-				return
+		if info.Netblock != netblock || info.ASN != asn {
-			}
+			t.Errorf("GetIPInfo(%s) = %s AS%d, want %s AS%d", ip, info.Netblock, info.ASN, netblock, asn)
-
+		}
-			if start != tt.wantStart {
-				t.Errorf("CalculateIPv4Range(%s) start = %d, want %d", tt.cidr, start, tt.wantStart)
-			}
-
-			if end != tt.wantEnd {
-				t.Errorf("CalculateIPv4Range(%s) end = %d, want %d", tt.cidr, end, tt.wantEnd)
-			}
-
-			// Verify that start <= end
-			if start > end {
-				t.Errorf("CalculateIPv4Range(%s) start (%d) > end (%d)", tt.cidr, start, end)
-			}
-
-			// Verify the range size matches the CIDR mask
-			if !tt.wantErr && tt.cidr != "" {
-				_, ipNet, _ := net.ParseCIDR(tt.cidr)
-				if ipNet != nil {
-					ones, bits := ipNet.Mask.Size()
-					expectedSize := uint32(1) << uint(bits-ones)
-					actualSize := end - start + 1
-					if actualSize != expectedSize {
-						t.Errorf("CalculateIPv4Range(%s) range size = %d, want %d", tt.cidr, actualSize, expectedSize)
-					}
-				}
-			}
-		})
-	}
-}
-
-func TestIPv4RangeIntegration(t *testing.T) {
-	// Test that our functions work correctly together
-	tests := []struct {
-		name          string
-		cidr          string
-		testIPs       []string
-		shouldContain []bool
-	}{
-		{
-			name: "192.168.1.0/24",
-			cidr: "192.168.1.0/24",
-			testIPs: []string{
-				"192.168.1.0",
-				"192.168.1.1",
-				"192.168.1.255",
-				"192.168.0.255",
-				"192.168.2.0",
-			},
-			shouldContain: []bool{true, true, true, false, false},
-		},
-		{
-			name: "10.0.0.0/8",
-			cidr: "10.0.0.0/8",
-			testIPs: []string{
-				"10.0.0.0",
-				"10.255.255.255",
-				"10.1.2.3",
-				"9.255.255.255",
-				"11.0.0.0",
-			},
-			shouldContain: []bool{true, true, true, false, false},
-		},
-		{
-			name: "172.16.0.0/12",
-			cidr: "172.16.0.0/12",
-			testIPs: []string{
-				"172.16.0.0",
-				"172.31.255.255",
-				"172.20.1.1",
-				"172.15.255.255",
-				"172.32.0.0",
-			},
-			shouldContain: []bool{true, true, true, false, false},
-		},
 	}
 
-	for _, tt := range tests {
+	// Default and very short routes, and routes one bit shorter than each limit.
-		t.Run(tt.name, func(t *testing.T) {
+	announce("2000::/3", 64500)
-			start, end, err := CalculateIPv4Range(tt.cidr)
+	announce("0.0.0.0/0", 64500)
-			if err != nil {
+	announce("2000::/15", 64500)
-				t.Fatalf("Failed to calculate range for %s: %v", tt.cidr, err)
+	announce("192.0.0.0/7", 64500)
-			}
 
-			for i, testIP := range tt.testIPs {
+	for _, ip := range []string{"2001:db8::1", "192.0.2.1"} {
-				ip := net.ParseIP(testIP)
+		if info, err := db.GetIPInfo(ip); !errors.Is(err, ErrNoRoute) {
-				if ip == nil {
+			t.Errorf("GetIPInfo(%s) with only short routes = %+v, %v; want ErrNoRoute", ip, info, err)
-					t.Fatalf("Failed to parse test IP: %s", testIP)
+		}
-				}
-
-				ipUint := ipToUint32(ip)
-				contained := ipUint >= start && ipUint <= end
-
-				if contained != tt.shouldContain[i] {
-					t.Errorf("IP %s in range %s: got %v, want %v", testIP, tt.cidr, contained, tt.shouldContain[i])
-				}
-			}
-		})
 	}
+
+	// Routes of exactly the minimum length are used.
+	announce("2001::/16", 64501)
+	announce("192.0.0.0/8", 64501)
+
+	lookup("2001:db8::1", "2001::/16", 64501)
+	lookup("192.0.2.1", "192.0.0.0/8", 64501)
+
+	announce("2001:db8::/32", 64502)
+
+	lookup("2001:db8::1", "2001:db8::/32", 64502)
 }
 
 // TestConnectionPoolPragmas holds several pooled connections open at once and
 // checks each one carries the per-connection settings from the DSN, plus the
-// process-wide hard heap limit.
+// process-wide hard heap limit, and sees the new file with auto_vacuum
+// incremental.
 func TestConnectionPoolPragmas(t *testing.T) {
 	cfg := &config.Config{StateDir: t.TempDir()}
 
@@ -372,6 +296,74 @@ func TestConnectionPoolPragmas(t *testing.T) {
 		if hardHeapLimit != sqliteHardHeapLimitBytes {
 			t.Errorf("conn %d: hard_heap_limit = %d, want %d", i, hardHeapLimit, sqliteHardHeapLimitBytes)
 		}
+
+		var autoVacuum int
+		if err := c.QueryRowContext(ctx, "PRAGMA auto_vacuum").Scan(&autoVacuum); err != nil {
+			t.Fatalf("conn %d: failed to read auto_vacuum: %v", i, err)
+		}
+		if autoVacuum != autoVacuumIncremental {
+			t.Errorf("conn %d: auto_vacuum = %d, want %d (incremental)", i, autoVacuum, autoVacuumIncremental)
+		}
+	}
+}
+
+// TestVacuumReturnsFreePages checks that after routes are deleted from a new
+// database, one Vacuum call returns every page they used. The deletes leave
+// fewer free pages than the 1000 Vacuum frees per call, so none may remain.
+// With auto_vacuum off (issue https://git.eeqj.de/sneak/routewatch/issues/43)
+// the free pages stayed in the file, and with the PRAGMA run by ExecContext
+// Vacuum freed only one page per call.
+func TestVacuumReturnsFreePages(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ctx := context.Background()
+	ts := time.Now().UTC()
+	const asn = 64500
+
+	routes := make([]*LiveRoute, 0, vacuumTestRoutes)
+	deletions := make([]LiveRouteDeletion, 0, vacuumTestRoutes)
+	for i := range vacuumTestRoutes {
+		route := mkV4Route(t, fmt.Sprintf("10.%d.%d.0/24", i/256, i%256), asn, ts)
+		routes = append(routes, route)
+		deletions = append(deletions, LiveRouteDeletion{
+			Prefix:    route.Prefix,
+			OriginASN: asn,
+			PeerIP:    route.PeerIP,
+			IPVersion: ipVersionV4,
+		})
+	}
+
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("failed to write routes: %v", err)
+	}
+	if err := db.DeleteLiveRouteBatch(deletions); err != nil {
+		t.Fatalf("failed to delete routes: %v", err)
+	}
+
+	var before int
+	if err := db.db.QueryRowContext(ctx, "PRAGMA freelist_count").Scan(&before); err != nil {
+		t.Fatalf("failed to read freelist_count: %v", err)
+	}
+	if before == 0 {
+		t.Fatalf("no free pages after deleting %d routes", vacuumTestRoutes)
+	}
+
+	if err := db.Vacuum(ctx); err != nil {
+		t.Fatalf("Vacuum failed: %v", err)
+	}
+
+	var after int
+	if err := db.db.QueryRowContext(ctx, "PRAGMA freelist_count").Scan(&after); err != nil {
+		t.Fatalf("failed to read freelist_count: %v", err)
+	}
+	if after != 0 {
+		t.Errorf("free pages after Vacuum = %d of %d, want 0", after, before)
 	}
 }
 
@@ -425,20 +417,280 @@ func TestBatchWriteDuringCheckpoint(t *testing.T) {
 	wg.Wait()
 }
 
-func BenchmarkIPToUint32(b *testing.B) {
+// TestGetRandomPrefixesByLengthListsPrefixOnce stores one prefix of each
-	ip := net.ParseIP("192.168.1.1")
+// family learned from two peers in the same second and checks that the
-	b.ResetTimer()
+// prefix-length sample lists each of them once.
+func TestGetRandomPrefixesByLengthListsPrefixOnce(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
 
-	for i := 0; i < b.N; i++ {
+	db, err := New(cfg, logger.New())
-		_ = ipToUint32(ip)
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	v4First := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	v4Second := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	v4Second.PeerIP = secondV4PeerIP
+	v6First := mkV6Route("2001:db8::/32", 64500, ts)
+	v6Second := mkV6Route("2001:db8::/32", 64500, ts)
+	v6Second.PeerIP = secondV6PeerIP
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{v4First, v4Second, v6First, v6Second}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	for _, want := range []*LiveRoute{v4First, v6First} {
+		routes, err := db.GetRandomPrefixesByLength(want.MaskLength, want.IPVersion, sampleLimit)
+		if err != nil {
+			t.Fatalf("GetRandomPrefixesByLength(%d, %d): %v", want.MaskLength, want.IPVersion, err)
+		}
+		if len(routes) != 1 || routes[0].Prefix != want.Prefix {
+			t.Errorf("GetRandomPrefixesByLength(%d, %d) = %+v, want %s listed once",
+				want.MaskLength, want.IPVersion, routes, want.Prefix)
+		}
 	}
 }
 
-func BenchmarkCalculateIPv4Range(b *testing.B) {
+// TestGetRandomPrefixesByLengthListsNewestRoute stores a prefix learned from
-	cidr := "192.168.0.0/16"
+// two peers at different times, the older route first, and checks that the
-	b.ResetTimer()
+// prefix-length sample lists the newer route.
+func TestGetRandomPrefixesByLengthListsNewestRoute(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
 
-	for i := 0; i < b.N; i++ {
+	db, err := New(cfg, logger.New())
-		_, _, _ = CalculateIPv4Range(cidr)
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	older := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	olderRoute := mkV4Route(t, "198.51.100.0/24", 64500, older)
+	newerRoute := mkV4Route(t, "198.51.100.0/24", 64500, older.Add(time.Minute))
+	newerRoute.PeerIP = secondV4PeerIP
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{olderRoute, newerRoute}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	routes, err := db.GetRandomPrefixesByLength(newerRoute.MaskLength, ipVersionV4, sampleLimit)
+	if err != nil {
+		t.Fatalf("GetRandomPrefixesByLength: %v", err)
+	}
+	if len(routes) != 1 || routes[0].PeerIP != newerRoute.PeerIP {
+		t.Errorf("GetRandomPrefixesByLength = %+v, want only the newer route, from %s",
+			routes, newerRoute.PeerIP)
+	}
+}
+
+// TestGetRandomPrefixesByLengthSamplesManyPrefixes stores more /24 prefixes
+// than the sample asks for, each followed by a /16 so that routes of another
+// mask length sit between the /24 rows, and each prefix learned from two peers
+// in the same second. It checks both the sample the pages get, picked from the
+// shuffled prefixes, and the random lookups used for a mask length with more
+// than maxPrefixesToShuffle prefixes.
+func TestGetRandomPrefixesByLengthSamplesManyPrefixes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	var routes []*LiveRoute
+	var stored []string
+	for i := range sampledPrefixCount {
+		prefix := fmt.Sprintf("10.0.%d.0/24", i)
+		for _, p := range []string{prefix, fmt.Sprintf("172.%d.0.0/16", i)} {
+			first := mkV4Route(t, p, 64500, ts)
+			second := mkV4Route(t, p, 64500, ts)
+			second.PeerIP = secondV4PeerIP
+			routes = append(routes, first, second)
+		}
+		stored = append(stored, prefix)
+	}
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	slices.Sort(stored)
+
+	maskLength := routes[0].MaskLength
+	sample, err := db.GetRandomPrefixesByLength(maskLength, ipVersionV4, sampleLimit)
+	if err != nil {
+		t.Fatalf("GetRandomPrefixesByLength: %v", err)
+	}
+	var listed []string
+	for _, route := range sample {
+		listed = append(listed, route.Prefix)
+	}
+	checkPrefixSample(t, "GetRandomPrefixesByLength", listed, stored)
+
+	looked, err := db.samplePrefixesByLength(
+		context.Background(), liveRoutesV4Table, maskLength, sampleLimit, nil)
+	if err != nil {
+		t.Fatalf("samplePrefixesByLength: %v", err)
+	}
+	checkPrefixSample(t, "samplePrefixesByLength", looked, stored)
+}
+
+// checkPrefixSample checks that sample lists sampleLimit prefixes, each once
+// and each one of stored, and that it is not just the first sampleLimit of
+// stored, which is in text order.
+func checkPrefixSample(t *testing.T, name string, sample, stored []string) {
+	t.Helper()
+
+	if len(sample) != sampleLimit {
+		t.Errorf("%s listed %d prefixes, want %d", name, len(sample), sampleLimit)
+	}
+	listed := make(map[string]bool)
+	for _, prefix := range sample {
+		if listed[prefix] || !slices.Contains(stored, prefix) {
+			t.Errorf("%s lists %s twice or it is not one of the /24s", name, prefix)
+		}
+		listed[prefix] = true
+	}
+	if slices.Equal(slices.Sorted(slices.Values(sample)), stored[:sampleLimit]) {
+		t.Errorf("%s lists just the first %d prefixes in text order", name, sampleLimit)
+	}
+}
+
+// TestGetRandomPrefixesByLengthSamplesPastFirstPrefixes stores more prefixes
+// of one mask length than maxPrefixesToShuffle. It checks that
+// firstPrefixesByLength reads just the first maxPrefixesToShuffle+1 in text
+// order, and that the sample lists the prefixes asked for, each once, some of
+// them after those first ones, which only the random lookups reach.
+func TestGetRandomPrefixesByLengthSamplesPastFirstPrefixes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	var routes []*LiveRoute
+	var stored []string
+	for i := range largeSamplePrefixCount {
+		route := mkV6Route(fmt.Sprintf("2001:%x::/32", i), 64500, ts)
+		routes = append(routes, route)
+		stored = append(stored, route.Prefix)
+	}
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	slices.Sort(stored)
+	firstPrefixes := stored[:maxPrefixesToShuffle+1]
+
+	maskLength := routes[0].MaskLength
+	read, err := db.firstPrefixesByLength(
+		context.Background(), liveRoutesV6Table, maskLength, maxPrefixesToShuffle+1)
+	if err != nil {
+		t.Fatalf("firstPrefixesByLength: %v", err)
+	}
+	if !slices.Equal(read, firstPrefixes) {
+		t.Errorf("firstPrefixesByLength read %d prefixes, want the first %d in text order",
+			len(read), len(firstPrefixes))
+	}
+
+	sample, err := db.GetRandomPrefixesByLength(maskLength, ipVersionV6, largeSampleLimit)
+	if err != nil {
+		t.Fatalf("GetRandomPrefixesByLength: %v", err)
+	}
+	if len(sample) != largeSampleLimit {
+		t.Errorf("GetRandomPrefixesByLength listed %d prefixes, want %d", len(sample), largeSampleLimit)
+	}
+	listed := make(map[string]bool)
+	pastFirstPrefixes := 0
+	for _, route := range sample {
+		if listed[route.Prefix] {
+			t.Errorf("GetRandomPrefixesByLength lists %s twice", route.Prefix)
+		}
+		listed[route.Prefix] = true
+		if !slices.Contains(firstPrefixes, route.Prefix) {
+			pastFirstPrefixes++
+		}
+	}
+	if pastFirstPrefixes == 0 {
+		t.Errorf("GetRandomPrefixesByLength lists only prefixes among the first %d in text order",
+			len(firstPrefixes))
+	}
+}
+
+// TestSamplePrefixesByLengthFillsUpFromShuffled stores a single prefix, so
+// every random lookup lands on it, and checks that the sample still lists
+// sampleLimit prefixes: the stored one, then the shuffled ones in order. The
+// stored prefix is the second shuffled one, so the fill-up must skip it.
+func TestSamplePrefixesByLengthFillsUpFromShuffled(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	var shuffled []string
+	for i := range sampledPrefixCount {
+		shuffled = append(shuffled, fmt.Sprintf("10.0.%d.0/24", i))
+	}
+	route := mkV4Route(t, shuffled[1], 64500, time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC))
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{route}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	sample, err := db.samplePrefixesByLength(
+		context.Background(), liveRoutesV4Table, route.MaskLength, sampleLimit, shuffled)
+	if err != nil {
+		t.Fatalf("samplePrefixesByLength: %v", err)
+	}
+	want := append([]string{shuffled[1], shuffled[0]}, shuffled[2:sampleLimit]...)
+	if !slices.Equal(sample, want) {
+		t.Errorf("samplePrefixesByLength = %v, want %v", sample, want)
+	}
+}
+
+// TestGetRandomPrefixesByLengthSkipsWithdrawnRoutes withdraws a route after
+// storing it, as can happen between the steps of GetRandomPrefixesByLength. It
+// checks that the random lookups pick nothing for a mask length with no route
+// left, and that only the prefixes that still have a route get their newest
+// route listed.
+func TestGetRandomPrefixesByLengthSkipsWithdrawnRoutes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	kept := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	withdrawn := mkV4Route(t, "172.16.0.0/16", 64500, ts)
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{kept, withdrawn}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	if err := db.DeleteLiveRoute(withdrawn.Prefix, withdrawn.OriginASN, withdrawn.PeerIP); err != nil {
+		t.Fatalf("DeleteLiveRoute: %v", err)
+	}
+
+	ctx := context.Background()
+	sample, err := db.samplePrefixesByLength(
+		ctx, liveRoutesV4Table, withdrawn.MaskLength, sampleLimit, []string{withdrawn.Prefix})
+	if err != nil {
+		t.Fatalf("samplePrefixesByLength: %v", err)
+	}
+	if len(sample) != 0 {
+		t.Errorf("samplePrefixesByLength = %v, want none for a mask length with no route left", sample)
+	}
+
+	routes, err := db.newestRouteOfEachPrefix(
+		ctx, liveRoutesV4Table, ipVersionV4, []string{withdrawn.Prefix, kept.Prefix})
+	if err != nil {
+		t.Fatalf("newestRouteOfEachPrefix: %v", err)
+	}
+	if len(routes) != 1 || routes[0].Prefix != kept.Prefix {
+		t.Errorf("newestRouteOfEachPrefix = %+v, want only the route of %s", routes, kept.Prefix)
 	}
 }

internal/database/interface.go

@@ -18,6 +18,8 @@ type Stats struct {
 	Peers                  int
 	FileSizeBytes          int64
 	LiveRoutes             int
+	IPv4Routes             int
+	IPv6Routes             int
 	OldestRoute            *time.Time
 	NewestRoute            *time.Time
 	IPv4PrefixDistribution []PrefixDistribution
@@ -61,8 +63,6 @@ type Store interface {
 	GetLiveRouteCountsContext(ctx context.Context) (ipv4Count, ipv6Count int, err error)
 
 	// IP lookup operations
-	GetASInfoForIP(ip string) (*ASInfo, error)
-	GetASInfoForIPContext(ctx context.Context, ip string) (*ASInfo, error)
 	GetIPInfo(ip string) (*IPInfo, error)
 	GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, error)
 
@@ -76,8 +76,8 @@ type Store interface {
 	GetASDetailsContext(ctx context.Context, asn int) (*ASN, []LiveRoute, error)
 	GetASPeers(asn int) ([]ASPeer, error)
 	GetASPeersContext(ctx context.Context, asn int) ([]ASPeer, error)
-	GetPrefixDetails(prefix string) ([]LiveRoute, error)
+	GetPrefixDetails(prefix string) ([]LiveRoute, map[int]ASN, error)
-	GetPrefixDetailsContext(ctx context.Context, prefix string) ([]LiveRoute, error)
+	GetPrefixDetailsContext(ctx context.Context, prefix string) ([]LiveRoute, map[int]ASN, error)
 	GetRandomPrefixesByLength(maskLength, ipVersion, limit int) ([]LiveRoute, error)
 	GetRandomPrefixesByLengthContext(ctx context.Context, maskLength, ipVersion, limit int) ([]LiveRoute, error)
 

internal/database/models.go

@@ -77,9 +77,6 @@ type LiveRoute struct {
 	ASPath      []int     `json:"as_path"`
 	NextHop     string    `json:"next_hop"`
 	LastUpdated time.Time `json:"last_updated"`
-	// IPv4 range fields for fast lookups (nil for IPv6)
-	V4IPStart *uint32 `json:"v4_ip_start,omitempty"`
-	V4IPEnd   *uint32 `json:"v4_ip_end,omitempty"`
 }
 
 // PrefixDistribution represents the distribution of prefixes by mask length
@@ -88,16 +85,6 @@ type PrefixDistribution struct {
 	Count      int `json:"count"`
 }
 
-// ASInfo represents AS information for an IP lookup (legacy format)
-type ASInfo struct {
-	ASN         int       `json:"asn"`
-	Handle      string    `json:"handle"`
-	Description string    `json:"description"`
-	Prefix      string    `json:"prefix"`
-	LastUpdated time.Time `json:"last_updated"`
-	Age         string    `json:"age"`
-}
-
 // IPInfo represents comprehensive IP information for the /ip endpoint
 type IPInfo struct {
 	IP         string   `json:"ip"`

internal/database/schema.sql

@@ -107,9 +107,6 @@ CREATE TABLE IF NOT EXISTS live_routes_v4 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	-- IPv4 range columns for fast lookups
-	ip_start INTEGER NOT NULL, -- Start of IPv4 range as 32-bit unsigned int
-	ip_end INTEGER NOT NULL,    -- End of IPv4 range as 32-bit unsigned int
 	UNIQUE(prefix, origin_asn, peer_ip)
 );
 
@@ -123,7 +120,6 @@ CREATE TABLE IF NOT EXISTS live_routes_v6 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	-- Note: IPv6 doesn't use integer range columns
 	UNIQUE(prefix, origin_asn, peer_ip)
 );
 
@@ -132,8 +128,6 @@ CREATE INDEX IF NOT EXISTS idx_live_routes_v4_prefix ON live_routes_v4(prefix);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_mask_length ON live_routes_v4(mask_length);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_origin_asn ON live_routes_v4(origin_asn);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_last_updated ON live_routes_v4(last_updated);
--- Indexes for IPv4 range queries
-CREATE INDEX IF NOT EXISTS idx_live_routes_v4_ip_range ON live_routes_v4(ip_start, ip_end);
 -- Index to optimize prefix distribution queries
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_mask_prefix ON live_routes_v4(mask_length, prefix);
 

internal/database/utils.go

@@ -1,6 +1,8 @@
 package database
 
 import (
+	"fmt"
+	"net"
 	"strings"
 
 	"github.com/google/uuid"
@@ -18,3 +20,14 @@ func detectIPVersion(prefix string) int {
 
 	return ipVersionV4
 }
+
+// prefixMaskLength returns the mask length of a prefix such as 192.0.2.0/24.
+func prefixMaskLength(prefix string) (int, error) {
+	_, network, err := net.ParseCIDR(prefix)
+	if err != nil {
+		return 0, fmt.Errorf("invalid prefix %s: %w", prefix, err)
+	}
+	maskLength, _ := network.Mask.Size()
+
+	return maskLength, nil
+}

internal/metrics/metrics.go

@@ -60,17 +60,22 @@ func New() *Tracker {
 	}
 }
 
-// SetConnected updates the connection status
+// SetConnected updates the connection status. Every connection after the
+// first counts one reconnect.
 func (t *Tracker) SetConnected(connected bool) {
-	wasConnected := t.isConnected.Swap(connected)
+	t.isConnected.Store(connected)
-	if connected {
+	if !connected {
-		t.mu.Lock()
+		return
-		t.connectedSince = time.Now()
+	}
-		t.mu.Unlock()
+
-		// Increment reconnect count (but not for the initial connection)
+	t.mu.Lock()
-		if wasConnected || t.reconnectCount.Load() > 0 {
+	// connectedSince is zero until the first connection.
-			t.reconnectCount.Add(1)
+	connectedBefore := !t.connectedSince.IsZero()
-		}
+	t.connectedSince = time.Now()
+	t.mu.Unlock()
+
+	if connectedBefore {
+		t.reconnectCount.Add(1)
 	}
 }
 

internal/metrics/metrics_test.go

@@ -0,0 +1,21 @@
+package metrics
+
+import "testing"
+
+// TestReconnectCount connects three times with a disconnect between each, the
+// way the streamer does, and expects every connection after the first to count
+// one reconnect.
+func TestReconnectCount(t *testing.T) {
+	tracker := New()
+
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+
+	const wantReconnects = 2
+	if got := tracker.GetReconnectCount(); got != wantReconnects {
+		t.Fatalf("reconnect count = %d, want %d", got, wantReconnects)
+	}
+}

internal/routewatch/app_integration_test.go

@@ -232,25 +232,6 @@ func (m *mockStore) GetLiveRouteCountsContext(ctx context.Context) (ipv4Count, i
 	return m.GetLiveRouteCounts()
 }
 
-// GetASInfoForIP mock implementation
-func (m *mockStore) GetASInfoForIP(ip string) (*database.ASInfo, error) {
-	// Simple mock - return a test AS
-	now := time.Now()
-	return &database.ASInfo{
-		ASN:         15169,
-		Handle:      "GOOGLE",
-		Description: "Google LLC",
-		Prefix:      "8.8.8.0/24",
-		LastUpdated: now.Add(-5 * time.Minute),
-		Age:         "5m0s",
-	}, nil
-}
-
-// GetASInfoForIPContext mock implementation with context support
-func (m *mockStore) GetASInfoForIPContext(ctx context.Context, ip string) (*database.ASInfo, error) {
-	return m.GetASInfoForIP(ip)
-}
-
 // GetASDetails mock implementation
 func (m *mockStore) GetASDetails(asn int) (*database.ASN, []database.LiveRoute, error) {
 	m.mu.Lock()
@@ -271,13 +252,15 @@ func (m *mockStore) GetASDetailsContext(ctx context.Context, asn int) (*database
 }
 
 // GetPrefixDetails mock implementation
-func (m *mockStore) GetPrefixDetails(prefix string) ([]database.LiveRoute, error) {
+func (m *mockStore) GetPrefixDetails(prefix string) ([]database.LiveRoute, map[int]database.ASN, error) {
 	// Return empty routes for now
-	return []database.LiveRoute{}, nil
+	return []database.LiveRoute{}, map[int]database.ASN{}, nil
 }
 
 // GetPrefixDetailsContext mock implementation with context support
-func (m *mockStore) GetPrefixDetailsContext(ctx context.Context, prefix string) ([]database.LiveRoute, error) {
+func (m *mockStore) GetPrefixDetailsContext(
+	ctx context.Context, prefix string,
+) ([]database.LiveRoute, map[int]database.ASN, error) {
 	return m.GetPrefixDetails(prefix)
 }
 
@@ -450,7 +433,7 @@ func TestRouteWatchLiveFeed(t *testing.T) {
 	}
 
 	// Create server
-	srv := server.New(mockDB, s, logger)
+	srv := server.New(mockDB, s, logger, cfg)
 
 	// Create RouteWatch with 5 second limit
 	deps := Dependencies{

internal/routewatch/cli.go

@@ -20,6 +20,8 @@ const (
 	debugInterval = 60 * time.Second
 	// bytesPerMB is bytes per megabyte
 	bytesPerMB = 1024 * 1024
+	// failureExitCode is the exit status when RouteWatch stops on an error
+	failureExitCode = 1
 )
 
 // logDebugStats logs goroutine count and memory usage
@@ -83,6 +85,9 @@ func CLIEntry() {
 					go func() {
 						if err := rw.Run(ctx); err != nil {
 							logger.Error("RouteWatch error", "error", err)
+							if err := shutdowner.Shutdown(fx.ExitCode(failureExitCode)); err != nil {
+								logger.Error("Failed to shutdown gracefully", "error", err)
+							}
 						}
 					}()
 

internal/routewatch/prefixhandler.go

@@ -2,6 +2,7 @@ package routewatch
 
 import (
 	"net"
+	"net/netip"
 	"strings"
 	"sync"
 	"time"
@@ -104,11 +105,28 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
-	// Process announcements
+	// Withdrawals are queued before announcements, because the batch keeps
+	// each peer's last update for a prefix: a message that lists a prefix as
+	// both withdrawn and announced keeps the route (RFC 4271 section 4.3).
+	for _, prefix := range msg.Withdrawals {
+		h.batch = append(h.batch, prefixUpdate{
+			prefix:      canonicalPrefix(prefix),
+			originASN:   originASN, // Use the originASN from path if available
+			peer:        msg.Peer,
+			messageType: "withdrawal",
+			timestamp:   timestamp,
+			path:        msg.Path,
+		})
+		// Record withdrawal in metrics
+		if h.metrics != nil {
+			h.metrics.RecordWithdrawal()
+		}
+	}
+
 	for _, announcement := range msg.Announcements {
 		for _, prefix := range announcement.Prefixes {
 			h.batch = append(h.batch, prefixUpdate{
-				prefix:      prefix,
+				prefix:      canonicalPrefix(prefix),
 				originASN:   originASN,
 				peer:        msg.Peer,
 				messageType: "announcement",
@@ -122,22 +140,6 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 		}
 	}
 
-	// Process withdrawals
-	for _, prefix := range msg.Withdrawals {
-		h.batch = append(h.batch, prefixUpdate{
-			prefix:      prefix,
-			originASN:   originASN, // Use the originASN from path if available
-			peer:        msg.Peer,
-			messageType: "withdrawal",
-			timestamp:   timestamp,
-			path:        msg.Path,
-		})
-		// Record withdrawal in metrics
-		if h.metrics != nil {
-			h.metrics.RecordWithdrawal()
-		}
-	}
-
 	// Check if we need to flush
 	if len(h.batch) >= prefixBatchSize {
 		h.flushBatchLocked()
@@ -178,14 +180,14 @@ func (h *PrefixHandler) flushBatchLocked() {
 	startTime := time.Now()
 	batchSize := len(h.batch)
 
-	// Group updates by prefix to deduplicate
+	// For each prefix and peer, keep only the last update the feed delivered.
-	// For each prefix, keep the latest update
+	// Updates from different peers for the same prefix are all kept. Arrival
-	prefixMap := make(map[string]prefixUpdate)
+	// order decides, not the timestamp: timestamps are whole seconds, so
+	// updates within one second carry the same one.
+	type prefixAndPeer struct{ prefix, peer string }
+	lastUpdates := make(map[prefixAndPeer]prefixUpdate)
 	for _, update := range h.batch {
-		key := update.prefix
+		lastUpdates[prefixAndPeer{update.prefix, update.peer}] = update
-		if existing, ok := prefixMap[key]; !ok || update.timestamp.After(existing.timestamp) {
-			prefixMap[key] = update
-		}
 	}
 
 	// Collect routes to upsert and delete
@@ -195,7 +197,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	// Collect unique prefixes to update
 	prefixesToUpdate := make(map[string]time.Time)
 
-	for _, update := range prefixMap {
+	for _, update := range lastUpdates {
 		// Track prefix for both announcements and withdrawals
 		if _, exists := prefixesToUpdate[update.prefix]; !exists || update.timestamp.After(prefixesToUpdate[update.prefix]) {
 			prefixesToUpdate[update.prefix] = update.timestamp
@@ -254,7 +256,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	elapsed := time.Since(startTime)
 	h.logger.Debug("Flushed prefix batch",
 		"batch_size", batchSize,
-		"unique_prefixes", len(prefixMap),
+		"unique_prefixes", len(prefixesToUpdate),
 		"success", successCount,
 		"duration_ms", elapsed.Milliseconds(),
 	)
@@ -264,6 +266,21 @@ func (h *PrefixHandler) flushBatchLocked() {
 	h.lastFlush = time.Now()
 }
 
+// canonicalPrefix returns prefix in the text form net/netip prints, the form
+// the IP lookup builds when it looks a prefix up. The feed sends IPv6
+// announcements compressed ("2001:db8::/32") but IPv6 withdrawals uncompressed
+// ("2001:db8:0:0:0:0:0:0/32"); stored as received, a withdrawal would not match
+// the route its announcement stored. A prefix that does not parse is returned
+// unchanged, and the batch flush reports it.
+func canonicalPrefix(prefix string) string {
+	p, err := netip.ParsePrefix(prefix)
+	if err != nil {
+		return prefix
+	}
+
+	return p.Masked().String()
+}
+
 // parseCIDR extracts the mask length and IP version from a prefix string
 func parseCIDR(prefix string) (maskLength int, ipVersion int, err error) {
 	_, ipNet, err := net.ParseCIDR(prefix)
@@ -315,20 +332,6 @@ func (h *PrefixHandler) processAnnouncement(_ *database.Prefix, update prefixUpd
 		LastUpdated: update.timestamp,
 	}
 
-	// For IPv4, calculate the IP range
-	if ipVersion == ipv4Version {
-		start, end, err := database.CalculateIPv4Range(update.prefix)
-		if err == nil {
-			liveRoute.V4IPStart = &start
-			liveRoute.V4IPEnd = &end
-		} else {
-			h.logger.Error("Failed to calculate IPv4 range",
-				"prefix", update.prefix,
-				"error", err,
-			)
-		}
-	}
-
 	if err := h.db.UpsertLiveRoute(liveRoute); err != nil {
 		h.logger.Error("Failed to upsert live route",
 			"prefix", update.prefix,
@@ -372,20 +375,6 @@ func (h *PrefixHandler) createLiveRoute(update prefixUpdate) *database.LiveRoute
 		LastUpdated: update.timestamp,
 	}
 
-	// For IPv4, calculate the IP range
-	if ipVersion == ipv4Version {
-		start, end, err := database.CalculateIPv4Range(update.prefix)
-		if err == nil {
-			liveRoute.V4IPStart = &start
-			liveRoute.V4IPEnd = &end
-		} else {
-			h.logger.Error("Failed to calculate IPv4 range",
-				"prefix", update.prefix,
-				"error", err,
-			)
-		}
-	}
-
 	return liveRoute
 }
 
@@ -425,20 +414,6 @@ func (h *PrefixHandler) processAnnouncementDirect(update prefixUpdate) {
 		LastUpdated: update.timestamp,
 	}
 
-	// For IPv4, calculate the IP range
-	if ipVersion == ipv4Version {
-		start, end, err := database.CalculateIPv4Range(update.prefix)
-		if err == nil {
-			liveRoute.V4IPStart = &start
-			liveRoute.V4IPEnd = &end
-		} else {
-			h.logger.Error("Failed to calculate IPv4 range",
-				"prefix", update.prefix,
-				"error", err,
-			)
-		}
-	}
-
 	if err := h.db.UpsertLiveRoute(liveRoute); err != nil {
 		h.logger.Error("Failed to upsert live route",
 			"prefix", update.prefix,

internal/routewatch/prefixhandler_test.go

@@ -0,0 +1,213 @@
+package routewatch
+
+import (
+	"context"
+	"errors"
+	"slices"
+	"testing"
+	"time"
+
+	"git.eeqj.de/sneak/routewatch/internal/config"
+	"git.eeqj.de/sneak/routewatch/internal/database"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"git.eeqj.de/sneak/routewatch/internal/ristypes"
+)
+
+const testPeerIP = "2001:db8:ffff::1"
+
+// TestPrefixHandlerStoresPrefixesTheIPLookupFinds runs announcements and
+// withdrawals through the prefix handler into a real database and looks the
+// addresses up. The prefixes are written the way the feed sends them: IPv6
+// announcements compressed, IPv6 withdrawals uncompressed. The withdrawal must
+// remove the route the announcement stored, and the IP lookup must find the
+// stored prefix.
+func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	// Built without NewPrefixHandler's flush timer, so the test flushes itself.
+	h := &PrefixHandler{db: db, logger: logger.New()}
+	flush := func() {
+		h.mu.Lock()
+		defer h.mu.Unlock()
+		h.flushBatchLocked()
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	h.HandleMessage(&ristypes.RISMessage{
+		Peer:            testPeerIP,
+		Path:            ristypes.ASPath{testASNA, testASNB},
+		ParsedTimestamp: ts,
+		Announcements: []ristypes.RISAnnouncement{{
+			NextHop:  testPeerIP,
+			Prefixes: []string{"2001:db8:1::/48", "192.0.2.0/24"},
+		}},
+	})
+	flush()
+
+	for ip, want := range map[string]string{
+		"2001:db8:1::1": "2001:db8:1::/48",
+		"192.0.2.1":     "192.0.2.0/24",
+	} {
+		info, err := db.GetIPInfo(ip)
+		if err != nil {
+			t.Fatalf("GetIPInfo(%s) after announcement: %v", ip, err)
+		}
+		if info.Netblock != want || info.ASN != testASNB {
+			t.Errorf("GetIPInfo(%s) = %s AS%d, want %s AS%d", ip, info.Netblock, info.ASN, want, testASNB)
+		}
+	}
+
+	h.HandleMessage(&ristypes.RISMessage{
+		Peer:            testPeerIP,
+		ParsedTimestamp: ts.Add(time.Minute),
+		Withdrawals:     []string{"2001:db8:1:0:0:0:0:0/48", "192.0.2.0/24"},
+	})
+	flush()
+
+	for _, ip := range []string{"2001:db8:1::1", "192.0.2.1"} {
+		if info, err := db.GetIPInfo(ip); !errors.Is(err, database.ErrNoRoute) {
+			t.Errorf("GetIPInfo(%s) after withdrawal = %+v, %v; want ErrNoRoute", ip, info, err)
+		}
+	}
+}
+
+// TestPrefixHandlerAppliesEachPeersLastUpdateInABatch runs updates for one
+// prefix through the prefix handler into a real database. All updates carry the
+// same timestamp, as updates within one second of the feed do. Within a batch,
+// the last update each peer sent for the prefix decides whether that peer has a
+// live route for it, and updates from different peers are all applied. The
+// in-memory route count and prefix distribution must match the route table.
+func TestPrefixHandlerAppliesEachPeersLastUpdateInABatch(t *testing.T) {
+	const (
+		prefix           = "198.51.100.0/24"
+		prefixMaskLength = 24
+		otherPeerIP      = "2001:db8:ffff::2"
+	)
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{
+			Peer:            peer,
+			Path:            ristypes.ASPath{testASNA, testASNB},
+			ParsedTimestamp: ts,
+			Announcements:   []ristypes.RISAnnouncement{{NextHop: peer, Prefixes: []string{prefix}}},
+		}
+	}
+	withdraw := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{Peer: peer, ParsedTimestamp: ts, Withdrawals: []string{prefix}}
+	}
+	announceAndWithdraw := func(peer string) *ristypes.RISMessage {
+		msg := announce(peer)
+		msg.Withdrawals = []string{prefix}
+
+		return msg
+	}
+
+	tests := []struct {
+		name string
+		// Each inner list is handled and then flushed as one batch.
+		batches [][]*ristypes.RISMessage
+		// The peers left with a live route for the prefix, by peer IP.
+		wantPeers []string
+	}{
+		{
+			name:      "two peers announce the prefix",
+			batches:   [][]*ristypes.RISMessage{{announce(testPeerIP), announce(otherPeerIP)}},
+			wantPeers: []string{testPeerIP, otherPeerIP},
+		},
+		{
+			name: "two peers withdraw the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announce(otherPeerIP)},
+				{withdraw(testPeerIP), withdraw(otherPeerIP)},
+			},
+		},
+		{
+			name:    "a peer announces and then withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{{announce(testPeerIP), withdraw(testPeerIP)}},
+		},
+		{
+			name: "a peer withdraws and then re-announces the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{withdraw(testPeerIP), announce(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+		{
+			// RFC 4271 section 4.3: the announcement counts, not the withdrawal.
+			name: "one message announces and withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announceAndWithdraw(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+			if err != nil {
+				t.Fatalf("failed to create database: %v", err)
+			}
+			defer func() { _ = db.Close() }()
+
+			// Built without NewPrefixHandler's flush timer, so the test flushes itself.
+			h := &PrefixHandler{db: db, logger: logger.New()}
+			for _, batch := range tt.batches {
+				for _, msg := range batch {
+					h.HandleMessage(msg)
+				}
+				h.mu.Lock()
+				h.flushBatchLocked()
+				h.mu.Unlock()
+			}
+
+			ctx := context.Background()
+			routes, _, err := db.GetPrefixDetailsContext(ctx, prefix)
+			if err != nil && !errors.Is(err, database.ErrNoRoute) {
+				t.Fatalf("GetPrefixDetailsContext: %v", err)
+			}
+			var gotPeers []string
+			for _, route := range routes {
+				gotPeers = append(gotPeers, route.PeerIP)
+			}
+			if !slices.Equal(gotPeers, tt.wantPeers) {
+				t.Errorf("peers with a live route for %s = %v, want %v", prefix, gotPeers, tt.wantPeers)
+			}
+
+			stats, err := db.GetStatsContext(ctx)
+			if err != nil {
+				t.Fatalf("GetStatsContext: %v", err)
+			}
+			tableRoutesV4, _, err := db.GetLiveRouteCountsContext(ctx)
+			if err != nil {
+				t.Fatalf("GetLiveRouteCountsContext: %v", err)
+			}
+			if stats.IPv4Routes != len(tt.wantPeers) || tableRoutesV4 != len(tt.wantPeers) {
+				t.Errorf("IPv4 live routes: %d in memory, %d in the table, want %d",
+					stats.IPv4Routes, tableRoutesV4, len(tt.wantPeers))
+			}
+
+			var wantDistribution []database.PrefixDistribution
+			if len(tt.wantPeers) > 0 {
+				wantDistribution = []database.PrefixDistribution{{MaskLength: prefixMaskLength, Count: 1}}
+			}
+			queryDistributionV4, _, err := db.GetPrefixDistributionContext(ctx)
+			if err != nil {
+				t.Fatalf("GetPrefixDistributionContext: %v", err)
+			}
+			if !slices.Equal(stats.IPv4PrefixDistribution, wantDistribution) ||
+				!slices.Equal(queryDistributionV4, wantDistribution) {
+				t.Errorf("IPv4 prefix distribution: %v in memory, %v from the table, want %v",
+					stats.IPv4PrefixDistribution, queryDistributionV4, wantDistribution)
+			}
+		})
+	}
+}

internal/server/handlers.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"net"
 	"net/http"
+	"net/netip"
 	"runtime"
 	"sort"
 	"strconv"
@@ -219,13 +220,6 @@ func (s *Server) handleStatusJSON() http.HandlerFunc {
 
 		const bitsPerMegabit = 1000000.0
 
-		// Get route counts from database
-		ipv4Routes, ipv6Routes, err := s.db.GetLiveRouteCountsContext(ctx)
-		if err != nil {
-			s.logger.Warn("Failed to get live route counts", "error", err)
-			// Continue with zero counts
-		}
-
 		// Get route update metrics
 		routeMetrics := s.streamer.GetMetricsTracker().GetRouteMetrics()
 
@@ -259,8 +253,8 @@ func (s *Server) handleStatusJSON() http.HandlerFunc {
 			Peers:                  dbStats.Peers,
 			DatabaseSizeBytes:      dbStats.FileSizeBytes,
 			LiveRoutes:             dbStats.LiveRoutes,
-			IPv4Routes:             ipv4Routes,
+			IPv4Routes:             dbStats.IPv4Routes,
-			IPv6Routes:             ipv6Routes,
+			IPv6Routes:             dbStats.IPv6Routes,
 			OldestRoute:            dbStats.OldestRoute,
 			NewestRoute:            dbStats.NewestRoute,
 			IPv4UpdatesPerSec:      routeMetrics.IPv4UpdatesPerSec,
@@ -439,13 +433,6 @@ func (s *Server) handleStats() http.HandlerFunc {
 
 		const bitsPerMegabit = 1000000.0
 
-		// Get route counts from database
-		ipv4Routes, ipv6Routes, err := s.db.GetLiveRouteCountsContext(ctx)
-		if err != nil {
-			s.logger.Warn("Failed to get live route counts", "error", err)
-			// Continue with zero counts
-		}
-
 		// Get route update metrics
 		routeMetrics := s.streamer.GetMetricsTracker().GetRouteMetrics()
 
@@ -537,8 +524,8 @@ func (s *Server) handleStats() http.HandlerFunc {
 			Peers:                  dbStats.Peers,
 			DatabaseSizeBytes:      dbStats.FileSizeBytes,
 			LiveRoutes:             dbStats.LiveRoutes,
-			IPv4Routes:             ipv4Routes,
+			IPv4Routes:             dbStats.IPv4Routes,
-			IPv6Routes:             ipv6Routes,
+			IPv6Routes:             dbStats.IPv6Routes,
 			OldestRoute:            dbStats.OldestRoute,
 			NewestRoute:            dbStats.NewestRoute,
 			IPv4UpdatesPerSec:      routeMetrics.IPv4UpdatesPerSec,
@@ -762,10 +749,17 @@ func (s *Server) handlePrefixDetailJSON() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
+		// Look the prefix up in the text form net/netip prints, the form the
-		prefix := prefixParam + "/" + lenParam
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			writeJSONError(w, http.StatusBadRequest, err.Error())
 
-		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
+			return
+		}
+		prefix := parsed.Masked().String()
+
+		routes, _, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				writeJSONError(w, http.StatusNotFound, err.Error())
@@ -926,10 +920,17 @@ func (s *Server) handlePrefixDetail() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
+		// Look the prefix up in the text form net/netip prints, the form the
-		prefix := prefixParam + "/" + lenParam
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			http.Error(w, "Invalid prefix", http.StatusBadRequest)
 
-		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
+			return
+		}
+		prefix := parsed.Masked().String()
+
+		routes, originASNs, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				http.Error(w, "Prefix not found", http.StatusNotFound)
@@ -951,18 +952,10 @@ func (s *Server) handlePrefixDetail() http.HandlerFunc {
 		originMap := make(map[int]*ASNInfo)
 		for _, route := range routes {
 			if _, exists := originMap[route.OriginASN]; !exists {
-				// Get AS info from database
-				asInfo, _, _ := s.db.GetASDetailsContext(r.Context(), route.OriginASN)
-				handle := ""
-				description := ""
-				if asInfo != nil {
-					handle = asInfo.Handle
-					description = asInfo.Description
-				}
 				originMap[route.OriginASN] = &ASNInfo{
 					ASN:         route.OriginASN,
-					Handle:      handle,
+					Handle:      originASNs[route.OriginASN].Handle,
-					Description: description,
+					Description: originASNs[route.OriginASN].Description,
 					PeerCount:   0,
 				}
 			}

internal/server/handlers_test.go

@@ -2,16 +2,24 @@ package server
 
 import (
 	"context"
+	"encoding/json"
+	"errors"
+	"html"
 	"net/http"
 	"net/http/httptest"
 	"runtime"
+	"slices"
+	"strings"
 	"testing"
 	"time"
 
+	"git.eeqj.de/sneak/routewatch/internal/config"
 	"git.eeqj.de/sneak/routewatch/internal/database"
 	"git.eeqj.de/sneak/routewatch/internal/logger"
 	"git.eeqj.de/sneak/routewatch/internal/metrics"
 	"git.eeqj.de/sneak/routewatch/internal/streamer"
+	"git.eeqj.de/sneak/routewatch/pkg/asinfo"
+	"github.com/google/uuid"
 )
 
 // blockingStatsDB embeds database.Store (left nil) and overrides only
@@ -38,7 +46,7 @@ func (d blockingStatsDB) GetStatsContext(_ context.Context) (database.Stats, err
 func TestStatsHandlersDoNotLeakOnTimeout(t *testing.T) {
 	release := make(chan struct{})
 	db := blockingStatsDB{release: release}
-	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New())
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
 
 	handlers := map[string]http.HandlerFunc{
 		"status.json": s.handleStatusJSON(),
@@ -70,6 +78,198 @@ func TestStatsHandlersDoNotLeakOnTimeout(t *testing.T) {
 	}
 }
 
+// TestStatsHandlersAnswerFromMemory checks that both stats handlers answer 200
+// with the live route counts and the prefix distribution while the database is
+// closed, so that any query would fail: the request path reads them from
+// memory. The prefix distribution query it used to run read every live route
+// and, on a large database, took the whole 4-second deadline, so
+// /api/v1/stats answered 500 (https://git.eeqj.de/sneak/routewatch/issues/30).
+// The oldest and newest route times still come from one-row lookups at the ends
+// of an index; with the database closed they are left out of the answer.
+func TestStatsHandlersAnswerFromMemory(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("database.New: %v", err)
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := db.UpsertLiveRouteBatch([]*database.LiveRoute{
+		{
+			ID: uuid.New(), Prefix: "198.51.100.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: 64500, PeerIP: "192.0.2.1", ASPath: []int{64500}, NextHop: "192.0.2.1",
+			LastUpdated: ts,
+		},
+		{
+			ID: uuid.New(), Prefix: "2001:db8::/32", MaskLength: 32, IPVersion: 6,
+			OriginASN: 64501, PeerIP: "2001:db8::1", ASPath: []int{64501}, NextHop: "2001:db8::1",
+			LastUpdated: ts,
+		},
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	if err := db.Close(); err != nil {
+		t.Fatalf("Close: %v", err)
+	}
+
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+	handlers := map[string]http.HandlerFunc{
+		"status.json": s.handleStatusJSON(),
+		"stats":       s.handleStats(),
+	}
+
+	for name, handler := range handlers {
+		rec := httptest.NewRecorder()
+		handler(rec, httptest.NewRequest(http.MethodGet, "/", nil))
+		if rec.Code != http.StatusOK {
+			t.Errorf("%s: status %d, want %d; body %s", name, rec.Code, http.StatusOK, rec.Body)
+
+			continue
+		}
+
+		var body struct {
+			Data struct {
+				IPv4Routes             int                           `json:"ipv4_routes"`
+				IPv6Routes             int                           `json:"ipv6_routes"`
+				IPv4PrefixDistribution []database.PrefixDistribution `json:"ipv4_prefix_distribution"`
+				IPv6PrefixDistribution []database.PrefixDistribution `json:"ipv6_prefix_distribution"`
+			} `json:"data"`
+		}
+		if err := json.Unmarshal(rec.Body.Bytes(), &body); err != nil {
+			t.Fatalf("%s: decoding the answer: %v", name, err)
+		}
+
+		if body.Data.IPv4Routes != 1 || body.Data.IPv6Routes != 1 {
+			t.Errorf("%s: routes = (v4 %d, v6 %d), want (1, 1)", name, body.Data.IPv4Routes, body.Data.IPv6Routes)
+		}
+		wantV4 := []database.PrefixDistribution{{MaskLength: 24, Count: 1}}
+		if !slices.Equal(body.Data.IPv4PrefixDistribution, wantV4) {
+			t.Errorf("%s: IPv4 distribution = %v, want %v", name, body.Data.IPv4PrefixDistribution, wantV4)
+		}
+		wantV6 := []database.PrefixDistribution{{MaskLength: 32, Count: 1}}
+		if !slices.Equal(body.Data.IPv6PrefixDistribution, wantV6) {
+			t.Errorf("%s: IPv6 distribution = %v, want %v", name, body.Data.IPv6PrefixDistribution, wantV6)
+		}
+	}
+}
+
+// TestPrefixDetailHandlersParsePrefix checks that both prefix pages find a
+// stored route whatever valid text form the URL writes its prefix in, and
+// answer 400 for a prefix that does not parse. The tables store the form
+// net/netip prints. Before the fix the URL text was looked up as typed, so
+// "2001:DB8:0:0::/32" was not found although "2001:db8::/32" was stored, and
+// "foo/24" answered 500 (https://git.eeqj.de/sneak/routewatch/issues/71).
+func TestPrefixDetailHandlersParsePrefix(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("database.New: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := db.UpsertLiveRouteBatch([]*database.LiveRoute{
+		{
+			ID: uuid.New(), Prefix: "192.0.2.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: 64500, PeerIP: "198.51.100.1", ASPath: []int{64500}, NextHop: "198.51.100.1",
+			LastUpdated: ts,
+		},
+		{
+			ID: uuid.New(), Prefix: "2001:db8::/32", MaskLength: 32, IPVersion: 6,
+			OriginASN: 64501, PeerIP: "2001:db8::1", ASPath: []int{64501}, NextHop: "2001:db8::1",
+			LastUpdated: ts,
+		},
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	tests := []struct {
+		path string
+		want int
+	}{
+		{"/api/v1/prefix/foo/24", http.StatusBadRequest},
+		{"/prefix/foo/24", http.StatusBadRequest},
+		{"/api/v1/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/api/v1/prefix/192.0.2.1/24", http.StatusOK},
+		{"/prefix/192.0.2.1/24", http.StatusOK},
+	}
+
+	for _, tt := range tests {
+		rec := httptest.NewRecorder()
+		s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, tt.path, nil))
+		if rec.Code != tt.want {
+			t.Errorf("%s: status %d, want %d; body %s", tt.path, rec.Code, tt.want, rec.Body)
+		}
+	}
+}
+
+// failingASDetailsDB wraps a real store and fails every AS details lookup.
+type failingASDetailsDB struct {
+	database.Store
+}
+
+func (failingASDetailsDB) GetASDetailsContext(_ context.Context, _ int) (*database.ASN, []database.LiveRoute, error) {
+	return nil, nil, errors.New("AS details lookup failed")
+}
+
+// TestPrefixDetailShowsOriginNames checks that the prefix page shows each
+// origin AS's handle and description without looking up the AS's details,
+// which also read every route the AS originates. Before the fix the page did
+// that lookup once per origin, so with it failing the names were missing
+// (https://git.eeqj.de/sneak/routewatch/issues/67). The AS path column shows
+// handles too, so the test looks for the origin list's "(handle)" form.
+func TestPrefixDetailShowsOriginNames(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("database.New: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	origins := []int{15169, 13335}
+	for _, asn := range origins {
+		if _, err := db.GetOrCreateASN(asn, ts); err != nil {
+			t.Fatalf("GetOrCreateASN(%d): %v", asn, err)
+		}
+	}
+	if err := db.UpsertLiveRouteBatch([]*database.LiveRoute{
+		{
+			ID: uuid.New(), Prefix: "192.0.2.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: origins[0], PeerIP: "198.51.100.1", ASPath: []int{origins[0]}, NextHop: "198.51.100.1",
+			LastUpdated: ts,
+		},
+		{
+			ID: uuid.New(), Prefix: "192.0.2.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: origins[1], PeerIP: "198.51.100.2", ASPath: []int{origins[1]}, NextHop: "198.51.100.2",
+			LastUpdated: ts,
+		},
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	s := New(failingASDetailsDB{db}, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+	rec := httptest.NewRecorder()
+	s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/prefix/192.0.2.0/24", nil))
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status %d, want %d; body %s", rec.Code, http.StatusOK, rec.Body)
+	}
+
+	page := rec.Body.String()
+	for _, asn := range origins {
+		info, ok := asinfo.Get(asn)
+		if !ok {
+			t.Fatalf("AS%d is not in the AS info data", asn)
+		}
+		if want := "(" + html.EscapeString(info.Handle) + ")"; !strings.Contains(page, want) {
+			t.Errorf("page does not show AS%d's handle %q", asn, want)
+		}
+		if want := html.EscapeString(info.Description); !strings.Contains(page, want) {
+			t.Errorf("page does not show AS%d's description %q", asn, want)
+		}
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {

internal/server/server.go

@@ -3,10 +3,13 @@ package server
 
 import (
 	"context"
+	"fmt"
+	"net"
 	"net/http"
-	"os"
+	"strconv"
 	"time"
 
+	"git.eeqj.de/sneak/routewatch/internal/config"
 	"git.eeqj.de/sneak/routewatch/internal/database"
 	"git.eeqj.de/sneak/routewatch/internal/logger"
 	"git.eeqj.de/sneak/routewatch/internal/streamer"
@@ -33,16 +36,18 @@ type Server struct {
 	db         database.Store
 	streamer   *streamer.Streamer
 	logger     *logger.Logger
+	port       int
 	srv        *http.Server
 	asnFetcher ASNFetcher
 }
 
 // New creates a new HTTP server
-func New(db database.Store, streamer *streamer.Streamer, logger *logger.Logger) *Server {
+func New(db database.Store, streamer *streamer.Streamer, logger *logger.Logger, cfg *config.Config) *Server {
 	s := &Server{
 		db:       db,
 		streamer: streamer,
 		logger:   logger,
+		port:     cfg.Port,
 	}
 
 	s.setupRoutes()
@@ -50,14 +55,12 @@ func New(db database.Store, streamer *streamer.Streamer, logger *logger.Logger)
 	return s
 }
 
-// Start starts the HTTP server
+// Start listens on the configured port and serves HTTP in the background.
+// It returns an error if it cannot listen, for example because the port is
+// already in use.
 func (s *Server) Start() error {
-	port := os.Getenv("PORT")
-	if port == "" {
-		port = "8080"
-	}
-
 	const (
+		network           = "tcp"
 		readHeaderTimeout = 40 * time.Second
 		readTimeout       = 60 * time.Second
 		writeTimeout      = 60 * time.Second
@@ -65,7 +68,7 @@ func (s *Server) Start() error {
 	)
 
 	s.srv = &http.Server{
-		Addr:              ":" + port,
+		Addr:              ":" + strconv.Itoa(s.port),
 		Handler:           s.router,
 		ReadHeaderTimeout: readHeaderTimeout,
 		ReadTimeout:       readTimeout,
@@ -73,12 +76,17 @@ func (s *Server) Start() error {
 		IdleTimeout:       idleTimeout,
 	}
 
-	s.logger.Info("Starting HTTP server", "port", port, "addr", s.srv.Addr)
+	s.logger.Info("Starting HTTP server", "port", s.port, "addr", s.srv.Addr)
+
+	listener, err := net.Listen(network, s.srv.Addr)
+	if err != nil {
+		return fmt.Errorf("starting HTTP server: %w", err)
+	}
+
+	s.logger.Info("HTTP server listening", "addr", listener.Addr().String())
 
-	// Start in goroutine but log when actually listening
 	go func() {
-		s.logger.Info("HTTP server listening", "addr", s.srv.Addr)
+		if err := s.srv.Serve(listener); err != nil && err != http.ErrServerClosed {
-		if err := s.srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
 			s.logger.Error("HTTP server error", "error", err)
 		}
 	}()

internal/server/server_test.go

@@ -0,0 +1,33 @@
+package server
+
+import (
+	"context"
+	"net"
+	"testing"
+
+	"git.eeqj.de/sneak/routewatch/internal/config"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"git.eeqj.de/sneak/routewatch/internal/metrics"
+	"git.eeqj.de/sneak/routewatch/internal/streamer"
+)
+
+// TestStartFailsWhenPortIsHeld holds a port with a listener of its own and
+// checks that Start on that port returns an error. Before the fix Start
+// listened in a goroutine, only logged the failure and returned nil, so the
+// daemon kept running without its web server.
+func TestStartFailsWhenPortIsHeld(t *testing.T) {
+	held, err := net.Listen("tcp", ":0")
+	if err != nil {
+		t.Fatalf("failed to open the test listener: %v", err)
+	}
+	defer func() { _ = held.Close() }()
+
+	port := held.Addr().(*net.TCPAddr).Port
+	log := logger.New()
+	s := New(nil, streamer.New(log, metrics.New()), log, &config.Config{Port: port})
+	defer func() { _ = s.Stop(context.Background()) }()
+
+	if err := s.Start(); err == nil {
+		t.Fatalf("Start returned nil while port %d was held by another listener", port)
+	}
+}

internal/streamer/streamer.go

@@ -7,6 +7,7 @@ import (
 	"compress/gzip"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"math"
@@ -48,6 +49,7 @@ const (
 	metricsUpdateRate     = time.Second
 	minBackoffDelay       = 5 * time.Second
 	maxBackoffDelay       = 320 * time.Second
+	streamIdleTimeout     = 60 * time.Second // Reconnect when no line arrives for this long
 	metricsLogInterval    = 10 * time.Second
 	bytesPerKB            = 1024
 	bytesPerMB            = 1024 * 1024
@@ -58,6 +60,9 @@ const (
 	backpressureSlope     = 2.0 // Slope for linear drop probability increase
 )
 
+// errStreamEnded is returned by stream when RIS Live ends the response.
+var errStreamEnded = errors.New("RIS Live ended the stream")
+
 // MessageHandler defines the interface for processing RIS messages.
 // Implementations must specify which message types they want to receive,
 // how to process messages, and their desired queue capacity.
@@ -107,6 +112,8 @@ type Streamer struct {
 	logger       *logger.Logger
 	client       *http.Client
 	url          string
+	idleTimeout  time.Duration // Drop the connection when no line arrives for this long
+	minBackoff   time.Duration // First wait before reconnecting; doubles up to maxBackoffDelay
 	handlers     []*handlerInfo
 	rawHandler   RawMessageHandler
 	mu           sync.RWMutex
@@ -124,8 +131,10 @@ type Streamer struct {
 // The metrics tracker is used to record message counts, bytes received, and connection status.
 func New(logger *logger.Logger, metrics *metrics.Tracker) *Streamer {
 	return &Streamer{
-		logger: logger,
+		logger:      logger,
-		url:    risLiveURL,
+		url:         risLiveURL,
+		idleTimeout: streamIdleTimeout,
+		minBackoff:  minBackoffDelay,
 		client: &http.Client{
 			Timeout: 0, // No timeout for streaming
 			Transport: &http.Transport{
@@ -210,9 +219,14 @@ func (s *Streamer) Start() error {
 // the connection status in metrics. This method is safe to call multiple times.
 func (s *Streamer) Stop() {
 	s.mu.Lock()
-	if s.cancel != nil {
+	if s.cancel == nil {
-		s.cancel()
+		// Not started, or already stopped: closing the queues again would panic.
+		s.mu.Unlock()
+
+		return
 	}
+	s.cancel()
+	s.cancel = nil
 	// Close all handler queues to signal workers to stop
 	for _, info := range s.handlers {
 		close(info.queue)
@@ -400,7 +414,7 @@ func (s *Streamer) updateMetrics(messageBytes int) {
 
 // streamWithReconnect handles streaming with automatic reconnection and exponential backoff
 func (s *Streamer) streamWithReconnect(ctx context.Context) {
-	backoffDelay := minBackoffDelay
+	backoffDelay := s.minBackoff
 	consecutiveFailures := 0
 
 	for {
@@ -417,8 +431,9 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 		err := s.stream(ctx)
 		streamDuration := time.Since(startTime)
 
-		if err == nil {
+		// Only Stop ends the streamer. A stream that ends any other way,
-			// Clean exit (context cancelled)
+		// with or without an error, is a disconnect and is reconnected.
+		if ctx.Err() != nil {
 			return
 		}
 
@@ -429,17 +444,13 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 			"stream_duration", streamDuration)
 		s.metrics.SetConnected(false)
 
-		// Check if context is cancelled
+		// A stream that lasted more than 30 seconds resets the backoff. That
-		if ctx.Err() != nil {
+		// includes a stream dropped after streamIdleTimeout with no line, so
-			return
+		// a silent feed is retried after the first backoff delay every time.
-		}
-
-		// If we streamed for more than 30 seconds, reset the backoff
-		// This indicates we had a successful connection that received data
 		if streamDuration > 30*time.Second {
 			s.logger.Info("Resetting backoff delay due to successful connection",
 				"stream_duration", streamDuration)
-			backoffDelay = minBackoffDelay
+			backoffDelay = s.minBackoff
 			consecutiveFailures = 0
 		} else {
 			// Increment consecutive failures
@@ -472,7 +483,13 @@ func (s *Streamer) stream(ctx context.Context) error {
 	connCtx, connCancel := context.WithCancel(ctx)
 	defer connCancel()
 
-	req, err := http.NewRequestWithContext(ctx, "GET", s.url, nil)
+	// The idle timer drops the connection, by cancelling connCtx, when no
+	// line arrives for s.idleTimeout. Each line read starts it over. It
+	// also covers the wait for the server to answer.
+	idleTimer := time.AfterFunc(s.idleTimeout, connCancel)
+	defer idleTimer.Stop()
+
+	req, err := http.NewRequestWithContext(connCtx, "GET", s.url, nil)
 	if err != nil {
 		return fmt.Errorf("failed to create request: %w", err)
 	}
@@ -554,6 +571,8 @@ func (s *Streamer) stream(ctx context.Context) error {
 	scanner := bufio.NewScanner(reader)
 
 	for scanner.Scan() {
+		idleTimer.Reset(s.idleTimeout)
+
 		select {
 		case <-ctx.Done():
 			s.logger.Info("Stream stopped by context")
@@ -660,8 +679,15 @@ func (s *Streamer) stream(ctx context.Context) error {
 			continue
 		}
 
-		// Dispatch to interested handlers
+		// Dispatch to interested handlers. Stop cancels ctx and closes the
+		// queues under the write lock, so if ctx is cancelled here, under the
+		// read lock, the queues are closed and must not be sent to.
 		s.mu.RLock()
+		if ctx.Err() != nil {
+			s.mu.RUnlock()
+
+			return ctx.Err()
+		}
 		for _, info := range s.handlers {
 			if !info.handler.WantsMessage(msg.Type) {
 				continue
@@ -695,11 +721,19 @@ func (s *Streamer) stream(ctx context.Context) error {
 		s.mu.RUnlock()
 	}
 
+	if ctx.Err() != nil {
+		return ctx.Err()
+	}
+
+	if connCtx.Err() != nil {
+		return fmt.Errorf("no line received for %s", s.idleTimeout)
+	}
+
 	if err := scanner.Err(); err != nil {
 		return fmt.Errorf("scanner error: %w", err)
 	}
 
-	return nil
+	return errStreamEnded
 }
 
 // shouldDropForBackpressure determines if a message should be dropped based on queue utilization

internal/streamer/streamer_test.go

@@ -2,14 +2,18 @@ package streamer
 
 import (
 	"context"
+	"errors"
+	"io"
 	"net/http"
 	"net/http/httptest"
 	"runtime"
+	"sync/atomic"
 	"testing"
 	"time"
 
 	"git.eeqj.de/sneak/routewatch/internal/logger"
 	"git.eeqj.de/sneak/routewatch/internal/metrics"
+	"git.eeqj.de/sneak/routewatch/internal/ristypes"
 )
 
 func TestNewStreamer(t *testing.T) {
@@ -36,6 +40,14 @@ func TestNewStreamer(t *testing.T) {
 	if s.metrics != metricsTracker {
 		t.Error("metrics tracker not set correctly")
 	}
+
+	if s.idleTimeout != streamIdleTimeout {
+		t.Errorf("idleTimeout = %s, want %s", s.idleTimeout, streamIdleTimeout)
+	}
+
+	if s.minBackoff != minBackoffDelay {
+		t.Errorf("minBackoff = %s, want %s", s.minBackoff, minBackoffDelay)
+	}
 }
 
 // TestStreamDoesNotLeakTickersAcrossReconnects drives many short-lived
@@ -54,8 +66,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	// One warm-up connection so any persistent HTTP transport goroutine exists
 	// before we take the baseline.
-	if err := s.stream(context.Background()); err != nil {
+	if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
-		t.Fatalf("warm-up stream returned error: %v", err)
+		t.Fatalf("warm-up stream returned %v, want %v", err, errStreamEnded)
 	}
 	s.client.CloseIdleConnections()
 
@@ -63,8 +75,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	const reconnects = 20
 	for range reconnects {
-		if err := s.stream(context.Background()); err != nil {
+		if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
-			t.Fatalf("stream returned error: %v", err)
+			t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
 		}
 	}
 	s.client.CloseIdleConnections()
@@ -75,6 +87,168 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 	}
 }
 
+// updateLine is one UPDATE message as RIS Live sends it.
+const updateLine = `{"type":"ris_message","data":{"type":"UPDATE","peer":"192.0.2.1",` +
+	`"peer_asn":"64496","timestamp":1700000000}}` + "\n"
+
+// updateHandler wants UPDATE messages. For each one it handles it sends on
+// handled if there is room, and otherwise does nothing; a nil handled is
+// never sent on.
+type updateHandler struct {
+	handled chan struct{}
+}
+
+func (updateHandler) WantsMessage(messageType string) bool { return messageType == "UPDATE" }
+func (updateHandler) QueueCapacity() int                   { return 10 }
+
+func (h updateHandler) HandleMessage(*ristypes.RISMessage) {
+	select {
+	case h.handled <- struct{}{}:
+	default:
+	}
+}
+
+// TestStopBeforeMessageReachesQueues stops the streamer after the read loop
+// has checked for cancellation but before it hands the message to the handler
+// queues. That is the gap a stop from another goroutine can land in, and it
+// used to end in "send on closed channel". The raw handler runs in that gap on
+// the read loop itself, so calling Stop from it hits the gap every time.
+func TestStopBeforeMessageReachesQueues(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		_, _ = io.WriteString(w, updateLine)
+	}))
+	defer srv.Close()
+
+	s := New(logger.New(), metrics.New())
+	s.url = srv.URL
+	s.RegisterHandler(updateHandler{})
+	s.RegisterRawHandler(func(string) { s.Stop() })
+
+	// Start would run the stream in the background, where the test cannot
+	// wait for it. Setting cancel as Start does lets Stop cancel the stream
+	// run here instead.
+	ctx, cancel := context.WithCancel(context.Background())
+	s.cancel = cancel
+
+	if err := s.stream(ctx); !errors.Is(err, context.Canceled) {
+		t.Fatalf("stream returned %v, want %v", err, context.Canceled)
+	}
+
+	// A second Stop must not close the queues again.
+	s.Stop()
+}
+
+// TestReconnectWhenStreamEnds has the server end the first response
+// normally, as when RIS Live restarts. The streamer must treat that as a
+// disconnect and connect again.
+func TestReconnectWhenStreamEnds(t *testing.T) {
+	endResponse := func(http.ResponseWriter, *http.Request) {}
+	checkReconnect(t, endResponse, streamIdleTimeout)
+}
+
+// TestReconnectWhenStreamGoesIdle has the server hold the first response
+// open without sending anything. The streamer must drop it once the idle
+// timeout passes and connect again.
+func TestReconnectWhenStreamGoesIdle(t *testing.T) {
+	const idleTimeout = 50 * time.Millisecond
+	holdResponse := func(w http.ResponseWriter, r *http.Request) {
+		w.(http.Flusher).Flush()
+		<-r.Context().Done()
+	}
+	checkReconnect(t, holdResponse, idleTimeout)
+}
+
+// TestReconnectWhenServerNeverAnswers has the server accept the first
+// request and never answer it, not even with headers. The streamer must give
+// up on it once the idle timeout passes and connect again.
+func TestReconnectWhenServerNeverAnswers(t *testing.T) {
+	const idleTimeout = 50 * time.Millisecond
+	neverAnswer := func(_ http.ResponseWriter, r *http.Request) {
+		<-r.Context().Done()
+	}
+	checkReconnect(t, neverAnswer, idleTimeout)
+}
+
+// TestStreamNotDroppedWhileLinesArrive has the server send lines well within
+// the idle timeout for several times the timeout, then end the response. Each
+// line must start the idle timeout over, so the stream lasts until the server
+// ends it instead of being dropped as idle.
+func TestStreamNotDroppedWhileLinesArrive(t *testing.T) {
+	const (
+		idleTimeout  = 100 * time.Millisecond
+		lineInterval = idleTimeout / 10
+		sendFor      = 5 * idleTimeout
+	)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		deadline := time.Now().Add(sendFor)
+		for time.Now().Before(deadline) {
+			_, _ = io.WriteString(w, updateLine)
+			w.(http.Flusher).Flush()
+			time.Sleep(lineInterval)
+		}
+	}))
+	defer srv.Close()
+
+	s := New(logger.New(), metrics.New())
+	s.url = srv.URL
+	s.idleTimeout = idleTimeout
+
+	if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
+		t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
+	}
+}
+
+// checkReconnect runs a streamer against a fake RIS Live server. The first
+// response is served by firstResponse; every later one carries one UPDATE
+// line and then stays open. The streamer must come back for a second
+// response, must not be marked connected when it does, and must handle the
+// UPDATE in it.
+func checkReconnect(t *testing.T, firstResponse http.HandlerFunc, idleTimeout time.Duration) {
+	t.Helper()
+
+	const maxWait = 5 * time.Second
+
+	tracker := metrics.New()
+	connectedAtReconnect := make(chan bool, 1)
+	var requests atomic.Int32
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if requests.Add(1) == 1 {
+			firstResponse(w, r)
+
+			return
+		}
+		select {
+		case connectedAtReconnect <- tracker.IsConnected():
+		default:
+		}
+		_, _ = io.WriteString(w, updateLine)
+		w.(http.Flusher).Flush()
+		<-r.Context().Done()
+	}))
+	defer srv.Close()
+
+	s := New(logger.New(), tracker)
+	s.url = srv.URL
+	s.idleTimeout = idleTimeout
+	s.minBackoff = time.Millisecond
+	handler := updateHandler{handled: make(chan struct{}, 1)}
+	s.RegisterHandler(handler)
+	if err := s.Start(); err != nil {
+		t.Fatalf("Start returned error: %v", err)
+	}
+	defer s.Stop()
+
+	select {
+	case <-handler.handled:
+	case <-time.After(maxWait):
+		t.Fatalf("no message from a second response handled within %s", maxWait)
+	}
+
+	if <-connectedAtReconnect {
+		t.Error("streamer was still marked connected when it came back for a second response")
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {

internal/templates/prefix_detail.html

@@ -283,6 +283,7 @@
                 <div class="origin-item">
                     <a href="/as/{{.ASN}}" class="as-link">AS{{.ASN}}</a>
                     {{if .Handle}} ({{.Handle}}){{end}}
+                    {{if .Description}} {{.Description}}{{end}}
                     <span style="color: #7f8c8d; margin-left: 10px;">{{.PeerCount}} peer{{if ne .PeerCount 1}}s{{end}}</span>
                 </div>
                 {{end}}

internal/version/version.go

@@ -7,7 +7,8 @@ package version
 var (
 	// GitRevision is the git commit hash
 	GitRevision = "unknown"
-	// GitRevisionShort is the short git commit hash (7 chars)
+	// GitRevisionShort is the version the page footer shows: the tag or
+	// short commit hash from `git describe --tags --always`
 	GitRevisionShort = "unknown"
 )
 

script/docker

@@ -1,7 +1,8 @@
 #!/bin/sh
 # script/docker: build the Docker image tagged with the project name.
 # Identical in all repos; the tag comes from script/projectname.
-# Generic: needs no adaptation.
+# --no-cache because the gate phases the final stage depends on are RUN
+# steps, and a cached one is a check that did not run.
 set -eu
 
 SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd -P)"
@@ -9,7 +10,15 @@ ROOT="$(cd "$SCRIPT_DIR/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    docker build -t "$("$SCRIPT_DIR/projectname")" .
+    # Own line: a failing command substitution inside an argument does
+    # not trip `set -e`, so the inline form degrades silently to an
+    # empty constant. The VERSION build argument takes precedence over
+    # the version a build stage derives from the .git in the context.
+    version="$(git describe --tags --always --dirty 2>/dev/null || true)"
+    [ -n "$version" ] || version="unknown"
+    docker build --no-cache \
+        --build-arg VERSION="$version" \
+        -t "$("$SCRIPT_DIR/projectname")" .
 }
 
 main "$@"