Make the tests pass on a busy host (closes #240)

Tests that waited a few seconds for something to happen now wait 15 seconds.
The wait ends as soon as the condition holds, so a passing run takes no longer.

TestStreamNotDroppedWhileLinesArrive, the tests whose requests pass a 2- or
4-second limit in the server, and TestStreamDurationsCountSeconds run on a fake
clock, so a stall on the host cannot change their result.

The checkpoint tests give the writing connection a 30-second busy timeout, so
their limit can be 15 seconds and a checkpoint that waits for a read still
fails them.

Model: opus-5-5

TODO.md

@@ -53,6 +53,9 @@ problem the run finds.
 
 # Completed Steps
 
+- 2026-11-10: the tests no longer fail when a busy host stalls for several
+  seconds. Most such waits are now 15 seconds. The test of the stream's idle
+  timeout runs on a fake clock (closes #240)
 - 2026-11-10: each result of the IP lookup has a `flags` object with `is_tor`
   and `is_bogon`. The public IP lists are named `ip_lists` in the answer,
   `internal/iplist` in the code and `iplists` in the state directory. A list

cmd/asinfo-gen/main_test.go

@@ -12,7 +12,7 @@ import (
 
 // parseTimeLimit is how long a test waits for parseCSV to return before it
 // fails.
-const parseTimeLimit = 2 * time.Second
+const parseTimeLimit = 15 * time.Second
 
 // TestParseCSVReadsTodaysLayout parses an as.csv with the country code column
 // that the source added after the description. It expects each AS with its

internal/database/database_internal_test.go

@@ -72,17 +72,34 @@ const (
 	// lockTestPrefix is the prefix the batch writes in the checkpoint tests
 	// store.
 	lockTestPrefix = "192.0.2.0/24"
+	// longBusyTimeoutMs is the busy timeout setLongBusyTimeout gives the
+	// writing connection. A checkpoint that waits for an open read takes this
+	// long.
+	longBusyTimeoutMs = 30000
 	// readerCheckpointLimit is how long a checkpoint, and a batch write
-	// waiting for it, may take while a reader is open: well under the busy
-	// timeout, which a checkpoint that waits for the reader takes.
-	readerCheckpointLimit = time.Second
+	// waiting for it, may take while a reader is open: well under
+	// longBusyTimeoutMs, with room for a stall on a busy host.
+	readerCheckpointLimit = 15 * time.Second
 	// truncateCheckpointLimit is how long a checkpoint of a WAL file over
 	// sqliteWALSizeLimitBytes may take to give up while a reader is open: its
-	// wait for the reader plus a second, well under the busy timeout.
+	// wait for the reader plus readerCheckpointLimit.
 	truncateCheckpointLimit = truncateBusyTimeoutMs*time.Millisecond +
-		time.Second
+		readerCheckpointLimit
 )
 
+// setLongBusyTimeout gives the writing connection of db longBusyTimeoutMs, so
+// that a checkpoint that waits for an open read takes far longer than a stall
+// on a busy host.
+func setLongBusyTimeout(ctx context.Context, t *testing.T, db *Database) {
+	t.Helper()
+
+	_, err := db.writer.ExecContext(ctx,
+		fmt.Sprintf("PRAGMA busy_timeout=%d", longBusyTimeoutMs))
+	if err != nil {
+		t.Fatalf("failed to set the busy timeout: %v", err)
+	}
+}
+
 // TestConnectionPoolPragmas holds several pooled connections open at once and
 // checks each one carries the per-connection settings from the DSN, plus the
 // process-wide heap limits, and sees the new file with auto_vacuum
@@ -579,6 +596,8 @@ func TestCheckpointDoesNotWaitForReader(t *testing.T) {
 	ctx := context.Background()
 	ts := time.Now().UTC()
 
+	setLongBusyTimeout(ctx, t, db)
+
 	err = db.UpdatePrefixesBatch(map[string]time.Time{"198.51.100.0/24": ts})
 	if err != nil {
 		t.Fatalf("failed to store the prefix the reader reads: %v", err)
@@ -690,6 +709,7 @@ func TestCheckpointEmptiesWALOnceNoReadIsOpen(t *testing.T) {
 	}
 
 	fillWALPastLimit(ctx, t, db)
+	setLongBusyTimeout(ctx, t, db)
 
 	started := time.Now()
 

internal/iplist/iplist_internal_test.go

@@ -380,9 +380,13 @@ func TestRestartLoadsCopiesFromDisk(t *testing.T) {
 	second.Start(t.Context())
 	second.Stop(t.Context())
 
+	// The download takes its time after writing the file, and the restart
+	// reads the file's time. A busy host can put seconds between the two.
+	const maxUpdatedDifference = 15 * time.Second
+
 	listings := lookup(t, second, listedAddr)
 	if len(listings) != 1 ||
-		listings[0].Updated.Sub(downloaded).Abs() > time.Second {
+		listings[0].Updated.Sub(downloaded).Abs() > maxUpdatedDifference {
 		t.Errorf("after the restart, listings %+v, want one updated at %s",
 			listings, downloaded)
 	}

internal/server/handlers_internal_test.go

@@ -17,6 +17,7 @@ import (
 	"strings"
 	"sync/atomic"
 	"testing"
+	"testing/synctest"
 	"time"
 
 	"git.eeqj.de/sneak/routewatch/internal/config"
@@ -173,6 +174,9 @@ type tableStats struct {
 // carried 203.0.113.0/24, the two with kept routes withdrew it, so its origin
 // keeps no route. The oldest and newest routes are those of the other two
 // prefixes.
+//
+// The requests run in a synctest bubble, whose clock is fake, so a stall on
+// the host does not use up the handlers' 4-second limit.
 func TestStatsHandlersAnswerFromMemory(t *testing.T) {
 	t.Parallel()
 
@@ -218,7 +222,11 @@ func TestStatsHandlersAnswerFromMemory(t *testing.T) {
 
 	for name, handler := range handlers {
 		rec := httptest.NewRecorder()
-		handler(rec, httptest.NewRequestWithContext(t.Context(), http.MethodGet, "/", nil))
+
+		synctest.Test(t, func(t *testing.T) {
+			handler(rec,
+				httptest.NewRequestWithContext(t.Context(), http.MethodGet, "/", nil))
+		})
 
 		if rec.Code != http.StatusOK {
 			t.Errorf("%s: status %d, want %d; body %s",
@@ -253,15 +261,20 @@ func TestStatsHandlersAnswerFromMemory(t *testing.T) {
 // TestStatsAnswerCarriesChurnPerMinute checks that the stream object of the
 // /api/v1/stats answer holds route_churn_per_min as a whole number, and no
 // longer route_churn_per_sec, the moving average it replaces
-// (https://git.eeqj.de/sneak/routewatch/issues/160).
+// (https://git.eeqj.de/sneak/routewatch/issues/160). The request runs in a
+// synctest bubble, whose clock is fake, so a stall on the host does not use up
+// the handler's 4-second limit.
 func TestStatsAnswerCarriesChurnPerMinute(t *testing.T) {
 	t.Parallel()
 
 	s := newServerWithTestRoutes(t)
 
 	rec := httptest.NewRecorder()
-	s.handleStats()(rec,
-		httptest.NewRequestWithContext(t.Context(), http.MethodGet, statsPath, nil))
+
+	synctest.Test(t, func(t *testing.T) {
+		s.handleStats()(rec,
+			httptest.NewRequestWithContext(t.Context(), http.MethodGet, statsPath, nil))
+	})
 
 	if rec.Code != http.StatusOK {
 		t.Fatalf("status %d, want %d; body %s", rec.Code, http.StatusOK, rec.Body)
@@ -309,18 +322,26 @@ func (heldBackHandler) HeldBackCount() int                 { return heldBackMess
 
 // TestStatsAnswerGivesHowManyMessagesAHandlerHoldsBack registers a handler
 // that holds back messages. Its figures in /api/v1/stats, which the status
-// page shows, must give how many.
+// page shows, must give how many. The request runs in a synctest bubble, whose
+// clock is fake, so a stall on the host does not use up the handler's 4-second
+// limit.
 func TestStatsAnswerGivesHowManyMessagesAHandlerHoldsBack(t *testing.T) {
 	t.Parallel()
 
 	s := newServerWithTestRoutes(t)
 	s.streamer.RegisterHandler(heldBackHandler{})
 
+	var answer []byte
+
+	synctest.Test(t, func(t *testing.T) {
+		answer = get(t, s, statsPath).Body.Bytes()
+	})
+
 	var body struct {
 		Data StatsResponse `json:"data"`
 	}
 
-	err := json.Unmarshal(get(t, s, statsPath).Body.Bytes(), &body)
+	err := json.Unmarshal(answer, &body)
 	if err != nil {
 		t.Fatalf("decoding the answer: %v", err)
 	}
@@ -364,7 +385,9 @@ type unitStats struct {
 // names. They used to be strings with the unit in them, such as "1h18m34s" or
 // "1.1 GB" (https://git.eeqj.de/sneak/routewatch/issues/199). The stream has
 // never connected, so the uptime is 0. The memory use in /api/v1/stats is the
-// same number of bytes as its gc.heap_alloc_bytes.
+// same number of bytes as its gc.heap_alloc_bytes. The requests run in a
+// synctest bubble, whose clock is fake, so a stall on the host does not use up
+// the handlers' 4-second limit.
 func TestStatsAnswersGiveUnitsInFieldNames(t *testing.T) {
 	t.Parallel()
 
@@ -374,7 +397,11 @@ func TestStatsAnswersGiveUnitsInFieldNames(t *testing.T) {
 	s.SetASNFetcher(fixedIntervalASNFetcher{interval: fetchInterval})
 
 	for _, path := range []string{statusJSONPath, statsPath} {
-		answer := get(t, s, path).Body.Bytes()
+		var answer []byte
+
+		synctest.Test(t, func(t *testing.T) {
+			answer = get(t, s, path).Body.Bytes()
+		})
 
 		var body struct {
 			Data unitStats `json:"data"`
@@ -415,82 +442,89 @@ func TestStatsAnswersGiveUnitsInFieldNames(t *testing.T) {
 // connection_duration_seconds of /api/v1/stats is null while the stream is not
 // connected, and a whole number of seconds once it is. It used to be the
 // string "disconnected", or a Go duration such as "1h18m34s"
-// (https://git.eeqj.de/sneak/routewatch/issues/199).
+// (https://git.eeqj.de/sneak/routewatch/issues/199). The test runs in a
+// synctest bubble, whose clock is fake, so a stall on the host does not use up
+// the handler's 4-second limit.
 func TestStatsAnswerGivesConnectionDurationInSeconds(t *testing.T) {
 	t.Parallel()
 
 	s := newServerWithTestRoutes(t)
 
-	connectionDuration := func() json.RawMessage {
-		answer := get(t, s, statsPath).Body.Bytes()
+	synctest.Test(t, func(t *testing.T) {
+		connectionDuration := func() json.RawMessage {
+			answer := get(t, s, statsPath).Body.Bytes()
 
-		var body struct {
-			Data map[string]json.RawMessage `json:"data"`
+			var body struct {
+				Data map[string]json.RawMessage `json:"data"`
+			}
+
+			err := json.Unmarshal(answer, &body)
+			if err != nil {
+				t.Fatalf("decoding the answer: %v; answer %s", err, answer)
+			}
+
+			duration, ok := body.Data["connection_duration_seconds"]
+			if !ok {
+				t.Fatalf("no connection_duration_seconds: %s", answer)
+			}
+
+			return duration
 		}
 
-		err := json.Unmarshal(answer, &body)
-		if err != nil {
-			t.Fatalf("decoding the answer: %v; answer %s", err, answer)
+		disconnected := connectionDuration()
+		if string(disconnected) != "null" {
+			t.Errorf("disconnected: connection_duration_seconds = %s, want null",
+				disconnected)
 		}
 
-		duration, ok := body.Data["connection_duration_seconds"]
-		if !ok {
-			t.Fatalf("no connection_duration_seconds: %s", answer)
+		s.streamer.GetMetricsTracker().SetConnected(true)
+
+		var seconds int64
+
+		connected := connectionDuration()
+
+		err := json.Unmarshal(connected, &seconds)
+		if err != nil || seconds != 0 {
+			t.Errorf("just connected: connection_duration_seconds = %s, want 0",
+				connected)
 		}
-
-		return duration
-	}
-
-	disconnected := connectionDuration()
-	if string(disconnected) != "null" {
-		t.Errorf("disconnected: connection_duration_seconds = %s, want null",
-			disconnected)
-	}
-
-	s.streamer.GetMetricsTracker().SetConnected(true)
-
-	var seconds int64
-
-	connected := connectionDuration()
-
-	err := json.Unmarshal(connected, &seconds)
-	if err != nil || seconds != 0 {
-		t.Errorf("just connected: connection_duration_seconds = %s, want 0",
-			connected)
-	}
+	})
 }
 
 // TestStreamDurationsCountSeconds checks that the uptime and the connection
 // duration of a stream that connected 90 seconds ago are both 90. A stream
 // that has never connected has an uptime of 0, and one that is not connected
-// has no connection duration.
+// has no connection duration. The test runs in a synctest bubble, whose clock
+// is fake, so a stall on the host does not add to the 90 seconds.
 func TestStreamDurationsCountSeconds(t *testing.T) {
 	t.Parallel()
 
-	const connectedFor = 90 * time.Second
+	synctest.Test(t, func(t *testing.T) {
+		const connectedFor = 90 * time.Second
 
-	connectedSince := time.Now().Add(-connectedFor)
-	want := int64(connectedFor / time.Second)
+		connectedSince := time.Now().Add(-connectedFor)
+		want := int64(connectedFor / time.Second)
 
-	if got := uptimeSeconds(connectedSince); got != want {
-		t.Errorf("uptimeSeconds = %d, want %d", got, want)
-	}
+		if got := uptimeSeconds(connectedSince); got != want {
+			t.Errorf("uptimeSeconds = %d, want %d", got, want)
+		}
 
-	if got := uptimeSeconds(time.Time{}); got != 0 {
-		t.Errorf("uptimeSeconds before any connection = %d, want 0", got)
-	}
+		if got := uptimeSeconds(time.Time{}); got != 0 {
+			t.Errorf("uptimeSeconds before any connection = %d, want 0", got)
+		}
 
-	switch got := connectionDurationSeconds(true, connectedSince); {
-	case got == nil:
-		t.Errorf("connectionDurationSeconds = nil, want %d", want)
-	case *got != want:
-		t.Errorf("connectionDurationSeconds = %d, want %d", *got, want)
-	}
+		switch got := connectionDurationSeconds(true, connectedSince); {
+		case got == nil:
+			t.Errorf("connectionDurationSeconds = nil, want %d", want)
+		case *got != want:
+			t.Errorf("connectionDurationSeconds = %d, want %d", *got, want)
+		}
 
-	if got := connectionDurationSeconds(false, connectedSince); got != nil {
-		t.Errorf("connectionDurationSeconds while not connected = %d, want nil",
-			*got)
-	}
+		if got := connectionDurationSeconds(false, connectedSince); got != nil {
+			t.Errorf("connectionDurationSeconds while not connected = %d, want nil",
+				*got)
+		}
+	})
 }
 
 // newTestTable returns a routing table that holds a route for testPrefixV4,
@@ -1134,7 +1168,9 @@ func (failingDB) GetPrefixFirstSeen(
 // with the driver's error text. It also sends a prefix that does not parse.
 // No answer may carry the driver's or the prefix parser's text. Before the fix
 // the answers sent it to the client
-// (https://git.eeqj.de/sneak/routewatch/issues/108).
+// (https://git.eeqj.de/sneak/routewatch/issues/108). The requests run in a
+// synctest bubble, whose clock is fake, so a stall on the host does not use up
+// the stats handlers' 4-second limit.
 func TestJSONRoutesDoNotSendErrorDetails(t *testing.T) {
 	t.Parallel()
 
@@ -1155,8 +1191,11 @@ func TestJSONRoutesDoNotSendErrorDetails(t *testing.T) {
 
 	for _, tt := range tests {
 		rec := httptest.NewRecorder()
-		s.router.ServeHTTP(rec, httptest.NewRequestWithContext(
-			t.Context(), http.MethodGet, tt.path, nil))
+
+		synctest.Test(t, func(t *testing.T) {
+			s.router.ServeHTTP(rec, httptest.NewRequestWithContext(
+				t.Context(), http.MethodGet, tt.path, nil))
+		})
 
 		if rec.Code != tt.wantCode {
 			t.Errorf("%s: status %d, want %d; body %s",
@@ -1397,8 +1436,11 @@ func TestIPLookupKeepsRouteWhenPTRLookupDoesNotAnswer(t *testing.T) {
 	s := newServerWithTestRoutes(t)
 	s.resolver = &fakeResolver{ptrNeverAnswers: true}
 
-	// The request's time limit passes after the PTR lookup's.
-	ctx, cancel := context.WithTimeout(t.Context(), 2*ptrLookupTimeout)
+	// The request's time limit passes well after the PTR lookup's, so that a
+	// stall on a busy host does not use up the difference.
+	const requestTimeLimit = ptrLookupTimeout + 15*time.Second
+
+	ctx, cancel := context.WithTimeout(t.Context(), requestTimeLimit)
 	defer cancel()
 
 	rec := httptest.NewRecorder()
@@ -2287,9 +2329,13 @@ func settledGoroutineCount() int {
 	return prev
 }
 
-// waitForGoroutines waits until the goroutine count drops to target or below.
+// waitForGoroutines waits up to 15 seconds for the goroutine count to drop to
+// target or below.
 func waitForGoroutines(target int) bool {
-	for range 100 {
+	const maxWait = 15 * time.Second
+
+	deadline := time.Now().Add(maxWait)
+	for time.Now().Before(deadline) {
 		if runtime.NumGoroutine() <= target {
 			return true
 		}

internal/server/home_internal_test.go

@@ -76,7 +76,9 @@ func asLink(asn int) string {
 }
 
 // TestHomePageCardsShowTheTableFigures fills a routing table and the asns
-// table, works out the cards, and checks that each card shows its figure.
+// table, works out the cards, and checks that each card shows its figure. The
+// cards are worked out in a synctest bubble, whose clock is fake, so a stall on
+// the host does not use up the 4-second limit on their database read.
 func TestHomePageCardsShowTheTableFigures(t *testing.T) {
 	t.Parallel()
 
@@ -95,7 +97,9 @@ func TestHomePageCardsShowTheTableFigures(t *testing.T) {
 		t.Fatalf("GetOrCreateASNBatch: %v", err)
 	}
 
-	s.refreshHomeCards(t.Context())
+	synctest.Test(t, func(t *testing.T) {
+		s.refreshHomeCards(t.Context())
+	})
 
 	page := get(t, s, "/").Body.String()
 
@@ -279,7 +283,7 @@ func TestKeepHomeCardsFreshWorksThemOutEveryInterval(t *testing.T) {
 
 	const (
 		interval = 10 * time.Millisecond
-		maxWait  = 5 * time.Second
+		maxWait  = 15 * time.Second
 	)
 
 	table := emptyTable()
@@ -321,7 +325,7 @@ func TestKeepHomeCardsFreshWorksThemOutEveryInterval(t *testing.T) {
 func TestStartedServerWorksOutTheCards(t *testing.T) {
 	t.Parallel()
 
-	const maxWait = 5 * time.Second
+	const maxWait = 15 * time.Second
 
 	s := newServerWithTable(t, newCardsTable())
 

internal/server/middleware_internal_test.go

@@ -14,6 +14,7 @@ import (
 	"regexp"
 	"strings"
 	"testing"
+	"testing/synctest"
 	"time"
 
 	"git.eeqj.de/sneak/routewatch/internal/config"
@@ -163,6 +164,10 @@ func TestJSONAnswersPutNoUnitInAString(t *testing.T) {
 // router and the middleware write. Before the fix a panic in a handler, and a
 // method /status.json or the health check does not take, answered with an
 // empty body (https://git.eeqj.de/sneak/routewatch/issues/201).
+//
+// The requests run in a synctest bubble, whose clock is fake, so a stall on
+// the host does not use up the 2-second limit of the health check or the
+// 4-second limit of the stats handlers.
 func TestJSONAnswersCarryMeta(t *testing.T) {
 	t.Parallel()
 
@@ -181,6 +186,10 @@ func TestJSONAnswersCarryMeta(t *testing.T) {
 	s.router.Get(panicsPath, JSONValidationMiddleware(http.HandlerFunc(
 		func(http.ResponseWriter, *http.Request) { panic(errDiskIO) })).ServeHTTP)
 
+	failing := New(failingDB{}, newTestTable(),
+		streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+	failing.resolver = &fakeResolver{}
+
 	tests := []struct {
 		method   string
 		path     string
@@ -216,42 +225,40 @@ func TestJSONAnswersCarryMeta(t *testing.T) {
 		{http.MethodGet, panicsPath, http.StatusInternalServerError},
 	}
 
-	for _, tt := range tests {
-		checkAnswerCarriesMeta(t, s, httptest.NewRequestWithContext(
-			t.Context(), tt.method, tt.path, nil), tt.wantCode)
-	}
+	synctest.Test(t, func(t *testing.T) {
+		for _, tt := range tests {
+			checkAnswerCarriesMeta(t, s, httptest.NewRequestWithContext(
+				t.Context(), tt.method, tt.path, nil), tt.wantCode)
+		}
 
-	// A request whose time limit has already passed answers 504. The handlers
-	// write the answers of the first five routes, and the JSON check on the
-	// route writes the last.
-	pastTimeLimit, cancel := context.WithDeadline(t.Context(), time.Now())
-	defer cancel()
+		// A request whose time limit has already passed answers 504. The
+		// handlers write the answers of the first five routes, and the JSON
+		// check on the route writes the last.
+		pastTimeLimit, cancel := context.WithDeadline(t.Context(), time.Now())
+		defer cancel()
 
-	for _, path := range []string{
-		statusJSONPath, statsPath, "/ip/" + addrInTestPrefixV4,
-		"/api/v1/ip/" + addrInTestPrefixV4, asJSONPath, writesNothingPath,
-	} {
-		checkAnswerCarriesMeta(t, s, httptest.NewRequestWithContext(
-			pastTimeLimit, http.MethodGet, path, nil), http.StatusGatewayTimeout)
-	}
+		for _, path := range []string{
+			statusJSONPath, statsPath, "/ip/" + addrInTestPrefixV4,
+			"/api/v1/ip/" + addrInTestPrefixV4, asJSONPath, writesNothingPath,
+		} {
+			checkAnswerCarriesMeta(t, s, httptest.NewRequestWithContext(
+				pastTimeLimit, http.MethodGet, path, nil), http.StatusGatewayTimeout)
+		}
 
-	// With a store whose reads fail, the health check answers 503, and the
-	// handlers of the other routes 500.
-	failing := New(failingDB{}, newTestTable(),
-		streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
-	failing.resolver = &fakeResolver{}
-
-	checkAnswerCarriesMeta(t, failing, httptest.NewRequestWithContext(
-		t.Context(), http.MethodGet, healthCheckPath, nil),
-		http.StatusServiceUnavailable)
-
-	for _, path := range []string{
-		statusJSONPath, statsPath, asJSONPath,
-		"/api/v1/ip/" + addrInTestPrefixV4, "/ip/" + addrInTestPrefixV4,
-	} {
+		// With a store whose reads fail, the health check answers 503, and
+		// the handlers of the other routes 500.
 		checkAnswerCarriesMeta(t, failing, httptest.NewRequestWithContext(
-			t.Context(), http.MethodGet, path, nil), http.StatusInternalServerError)
-	}
+			t.Context(), http.MethodGet, healthCheckPath, nil),
+			http.StatusServiceUnavailable)
+
+		for _, path := range []string{
+			statusJSONPath, statsPath, asJSONPath,
+			"/api/v1/ip/" + addrInTestPrefixV4, "/ip/" + addrInTestPrefixV4,
+		} {
+			checkAnswerCarriesMeta(t, failing, httptest.NewRequestWithContext(
+				t.Context(), http.MethodGet, path, nil), http.StatusInternalServerError)
+		}
+	})
 }
 
 // checkAnswerCarriesMeta serves req through the server's router, and checks

internal/server/routes_internal_test.go

@@ -5,6 +5,7 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"testing"
+	"testing/synctest"
 )
 
 // TestRoutesAnswerHEAD sends HEAD for a page and JSON routes through the
@@ -12,15 +13,23 @@ import (
 // lookup of the caller's own address, are in the list because chi's GetHead
 // does not reach them. Before the fix every route answered HEAD with 405
 // (https://git.eeqj.de/sneak/routewatch/issues/111).
+//
+// The requests run in a synctest bubble, whose clock is fake, so a stall on
+// the host does not use up the stats handlers' 4-second limit. The resolver
+// is fake because no time limit would cut off a real DNS lookup in a bubble.
 func TestRoutesAnswerHEAD(t *testing.T) {
 	t.Parallel()
 
 	s := newServerWithTestRoutes(t)
+	s.resolver = &fakeResolver{}
 
 	for _, path := range []string{"/", statusJSONPath, statsPath, "/ip", "/ip/"} {
 		rec := httptest.NewRecorder()
-		s.router.ServeHTTP(rec, httptest.NewRequestWithContext(
-			t.Context(), http.MethodHead, path, nil))
+
+		synctest.Test(t, func(t *testing.T) {
+			s.router.ServeHTTP(rec, httptest.NewRequestWithContext(
+				t.Context(), http.MethodHead, path, nil))
+		})
 
 		if rec.Code != http.StatusOK {
 			t.Errorf("HEAD %s: status %d, want %d; body %s",

internal/server/server_internal_test.go

@@ -22,10 +22,10 @@ func TestStopClosesTheConnectionOfARequestStillRunning(t *testing.T) {
 		stopDeadline = 100 * time.Millisecond
 		// waitTimeout is how long the test waits for the request to reach
 		// the handler, and then for the client to see its connection closed.
-		waitTimeout = 2 * time.Second
+		waitTimeout = 15 * time.Second
 		// readHeaderTimeout bounds the test server's read of the request
 		// header, as every http.Server must.
-		readHeaderTimeout = time.Second
+		readHeaderTimeout = waitTimeout
 	)
 
 	release := make(chan struct{})

internal/streamer/streamer_internal_test.go

@@ -14,6 +14,7 @@ import (
 	"strings"
 	"sync/atomic"
 	"testing"
+	"testing/synctest"
 	"time"
 
 	"git.eeqj.de/sneak/routewatch/internal/logger"
@@ -398,7 +399,7 @@ func TestStopQueuesTheHeldStateMessages(t *testing.T) {
 	const (
 		held    = 2
 		lines   = held + 1
-		maxWait = 5 * time.Second
+		maxWait = 15 * time.Second
 	)
 
 	srv := httptest.NewServer(http.HandlerFunc(
@@ -478,15 +479,15 @@ func TestStopQueuesTheHeldStateMessages(t *testing.T) {
 // TestStopReturnsByItsDeadline stops a streamer whose handler has a full queue
 // and a STATE message held for it, and a worker that stays busy until the test
 // ends, as one does whose handler waits for a database write. Stop must return
-// within its deadline plus a small margin instead of waiting for room on the
-// queue and for the worker, and log as not handed over the held message and
-// every message on the queue.
+// within its deadline plus a margin instead of waiting for room on the queue
+// and for the worker, and log as not handed over the held message and every
+// message on the queue.
 func TestStopReturnsByItsDeadline(t *testing.T) {
 	t.Parallel()
 
 	const (
 		deadline = 100 * time.Millisecond
-		margin   = 2 * time.Second
+		margin   = 15 * time.Second
 	)
 
 	var logs bytes.Buffer
@@ -610,10 +611,34 @@ func TestReconnectWhenServerNeverAnswers(t *testing.T) {
 	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.
+// pipeTransport answers every request with a response whose body write fills.
+// The body fails when the request's context ends, as a connection's does.
+// write gets that context, so that it can end when the request does.
+type pipeTransport struct {
+	write func(context.Context, io.Writer)
+}
+
+func (p pipeTransport) RoundTrip(req *http.Request) (*http.Response, error) {
+	body, w := io.Pipe()
+
+	context.AfterFunc(req.Context(), func() {
+		_ = w.CloseWithError(req.Context().Err())
+	})
+
+	go func(ctx context.Context) {
+		p.write(ctx, w)
+
+		_ = w.Close()
+	}(req.Context())
+
+	return &http.Response{StatusCode: http.StatusOK, Body: body}, nil
+}
+
+// TestStreamNotDroppedWhileLinesArrive has a response send lines well within
+// the idle timeout for several times the timeout, then end. Each line must
+// start the idle timeout over, so the stream lasts until the response ends
+// instead of being dropped as idle. It runs in a synctest bubble, whose clock
+// is fake, so a stall on the host does not use up the timeout.
 func TestStreamNotDroppedWhileLinesArrive(t *testing.T) {
 	t.Parallel()
 
@@ -623,26 +648,35 @@ func TestStreamNotDroppedWhileLinesArrive(t *testing.T) {
 		sendFor      = 5 * idleTimeout
 	)
 
-	srv := httptest.NewServer(http.HandlerFunc(
-		func(w http.ResponseWriter, _ *http.Request) {
+	// The streamer is made outside the bubble. Its metrics start a goroutine
+	// that never returns, and a bubble waits for each of its goroutines to
+	// return.
+	s := New(logger.New(), metrics.New())
+	s.idleTimeout = idleTimeout
+	s.client = &http.Client{Transport: pipeTransport{
+		write: func(ctx context.Context, w io.Writer) {
 			deadline := time.Now().Add(sendFor)
 			for time.Now().Before(deadline) {
 				_, _ = io.WriteString(w, updateLine)
 
-				flush(w)
-				time.Sleep(lineInterval)
+				// The wait ends with the stream. A bubble's clock stops when
+				// the test function returns, and a goroutine then still
+				// asleep panics the test binary.
+				select {
+				case <-ctx.Done():
+					return
+				case <-time.After(lineInterval):
+				}
 			}
-		}))
-	defer srv.Close()
+		},
+	}}
 
-	s := New(logger.New(), metrics.New())
-	s.url = srv.URL
-	s.idleTimeout = idleTimeout
-
-	err := s.stream(context.Background())
-	if !errors.Is(err, errStreamEnded) {
-		t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
-	}
+	synctest.Test(t, func(t *testing.T) {
+		err := s.stream(context.Background())
+		if !errors.Is(err, errStreamEnded) {
+			t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
+		}
+	})
 }
 
 // TestStreamReadsLongLine has the server send one UPDATE line of about
@@ -706,7 +740,7 @@ func TestHandlerStatsReadWhileMessagesAreDropped(t *testing.T) {
 	const (
 		lines       = 5
 		wantDropped = lines - 2
-		maxWait     = 5 * time.Second
+		maxWait     = 15 * time.Second
 	)
 
 	srv := httptest.NewServer(http.HandlerFunc(
@@ -825,7 +859,7 @@ func checkReconnect(
 ) {
 	t.Helper()
 
-	const maxWait = 5 * time.Second
+	const maxWait = 15 * time.Second
 
 	tracker := metrics.New()
 	connectedAtReconnect := make(chan bool, 1)
@@ -896,9 +930,13 @@ func settledGoroutineCount() int {
 	return prev
 }
 
-// waitForGoroutines waits until the goroutine count drops to target or below.
+// waitForGoroutines waits up to 15 seconds for the goroutine count to drop to
+// target or below.
 func waitForGoroutines(target int) bool {
-	for range 100 {
+	const maxWait = 15 * time.Second
+
+	deadline := time.Now().Add(maxWait)
+	for time.Now().Before(deadline) {
 		if runtime.NumGoroutine() <= target {
 			return true
 		}