The binary make install builds retries a refused or unresolvable connection (closes #206)

Bun's `fetch`, which `~/bin/quak` runs under, rejects with a plain `Error`
that has no `cause` and a code of bun's own. Running the built binary against
local listeners gave `ConnectionRefused` for a refused connection and for a
failed TLS handshake, and `ConnectionRefused` or `FailedToOpenSocket` for a
name that does not resolve. Both codes join `TRANSPORT_CODES` and
`CONNECT_CODES` in `src/retry.ts`, so they are retried, replayed, and called
transient by `classify`. A reset already came as `ECONNRESET` and a deadline
as a `TimeoutError`; the tests pin every shape found.

Judgement call: a failed TLS handshake is replayed, as no request byte was
sent.
Not changed: a certificate bun rejects is not retried.

Model: opus-5-5

README.md

@@ -151,7 +151,9 @@ for each, separated by `, `. A file whose download failed that way is in
 `failures.json` with that error, as a transient failure. So is each file with no
 ML data cached when the ML data fetch failed that way, its error starting
 `ML data: `. The `~/bin/quak` that `make install` builds runs under bun, and
-prints bun's own wording for such a failure.
+prints bun's own wording for such a failure, with no reason after it, such as
+`quak: Unable to connect. Is the computer able to access the url?` for a refused
+connection. It records such a failure in `failures.json` as transient too.
 
 After a `1`, the next run fetches each missing original again, fetches the ML
 data that is not cached, and rebuilds the album links. An album quak could not
@@ -494,6 +496,17 @@ request is repeated only when repeating it could produce a different answer:
   declining to retry a transient one costs a single file that the next run picks
   up.
 
+The `~/bin/quak` that `make install` builds runs under bun, whose `fetch`
+rejects with a plain `Error` that has no `cause` and carries a code of bun's
+own. quak retries a refused connection, a name that does not resolve, a reset
+connection and a deadline under bun as it does under Node. Bun reports a refused
+connection, and a TLS handshake that failed, as `ConnectionRefused`. For a name
+that does not resolve it gives `ConnectionRefused` on some requests and
+`FailedToOpenSocket` on others. A connection reset or closed before the response
+ended is `ECONNRESET`, as under Node, and a deadline is the same `TimeoutError`.
+A certificate bun rejects, reported with a code such as
+`DEPTH_ZERO_SELF_SIGNED_CERT`, is not retried.
+
 Backoff is exponential with full jitter: the delay before retry _n_ is
 `random() * min(maxDelayMs, baseDelayMs * 2 ** (n - 1))`. The exponential term
 is the ceiling and the wait is drawn below it, so a client that lost many
@@ -541,17 +554,20 @@ second-factor attempts. They are retried only when every errno in the error's
 `cause` chain is one of the three that establish no TCP connection to the server
 ever existed, so no request byte can have been transmitted: `ENOTFOUND` and
 `EAI_AGAIN` (name resolution produced no address) and `ECONNREFUSED` (the peer
-refused the connection). A 5xx, a mid-flight reset and a deadline are all left
-to the caller, because each of them can happen after the server has already
-acted. These two do not follow redirects either: a redirect means the server
-already received the request, so it is reported as an error and not retried. The
-routing errnos `EHOSTUNREACH`, `ENETUNREACH` and `ENETDOWN` are excluded for the
-same reason, despite looking like connect-time failures: on Linux an ICMP
-unreachable arriving mid-flight, or a local interface going down after the
-request was written, delivers them on an already-established socket. They stay
-retryable for the idempotent calls. `putFile` is exempt: a presigned PUT stores
-one whole object at one key in one request, so replaying it has no partial state
-to damage.
+refused the connection). Under bun these failures are `ConnectionRefused` and
+`FailedToOpenSocket`. Bun also gives `ConnectionRefused` for a TLS handshake
+that failed; the TCP connection existed then, but the request is only sent once
+the handshake is done. A 5xx, a mid-flight reset and a deadline are all left to
+the caller, because each of them can happen after the server has already acted.
+These two do not follow redirects either: a redirect means the server already
+received the request, so it is reported as an error and not retried. The routing
+errnos `EHOSTUNREACH`, `ENETUNREACH` and `ENETDOWN` are excluded for the same
+reason, despite looking like connect-time failures: on Linux an ICMP unreachable
+arriving mid-flight, or a local interface going down after the request was
+written, delivers them on an already-established socket. They stay retryable for
+the idempotent calls. `putFile` is exempt: a presigned PUT stores one whole
+object at one key in one request, so replaying it has no partial state to
+damage.
 
 A download is retried as a whole — request, stream consumption, and decryption —
 because a socket reset after the response headers have arrived surfaces in the

TODO.md

@@ -17,11 +17,10 @@ pre-1.0
 
 # Next Step
 
-Five bugs are open: https://git.eeqj.de/sneak/quak/issues/197,
+Four bugs are open: https://git.eeqj.de/sneak/quak/issues/197,
 https://git.eeqj.de/sneak/quak/issues/198,
-https://git.eeqj.de/sneak/quak/issues/199,
-https://git.eeqj.de/sneak/quak/issues/205 and
-https://git.eeqj.de/sneak/quak/issues/206. The check of quak against a live Ente
+https://git.eeqj.de/sneak/quak/issues/199 and
+https://git.eeqj.de/sneak/quak/issues/205. The check of quak against a live Ente
 account, https://git.eeqj.de/sneak/quak/issues/187, needs an account only sneak
 has, so it waits on him.
 
@@ -30,6 +29,14 @@ declares one.
 
 # Completed Steps
 
+- 2026-10-16: The `~/bin/quak` that `make install` builds retries a refused
+  connection and a name that does not resolve, and the backup records either as
+  transient (issue 206). Bun's `fetch` reports them with codes of its own,
+  `ConnectionRefused` and `FailedToOpenSocket`, which `src/retry.ts` now counts
+  as transport failures that are safe to replay. Bun already gave `ECONNRESET`
+  for a reset and a `TimeoutError` for a deadline. The shapes were found by
+  running the binary against local listeners.
+
 - 2026-10-15: A network failure is printed and recorded with its reason, not as
   just `fetch failed` (issue 196). `describeError` in `src/errors.ts` gives an
   error's message followed by the message of each error on its `cause` chain,

src/retry.ts

@@ -58,6 +58,13 @@ export const resolveRetryOptions = (
 });
 
 // Transport failures: the request did not complete, for reasons below HTTP.
+//
+// The errnos are Node's. The binary `make install` builds runs under bun,
+// whose `fetch` rejects with a plain `Error` that has no `cause` and a `code`
+// of bun's own; `CONNECT_CODES` says what bun's two codes mean. For a
+// connection reset or closed before the response ended, bun gives Node's
+// `ECONNRESET`. These were found by running that binary, built with bun
+// 1.3.13, against listeners on 127.0.0.1.
 const TRANSPORT_CODES = new Set([
     "ECONNRESET",
     "ECONNABORTED",
@@ -70,12 +77,19 @@ const TRANSPORT_CODES = new Set([
     "ENETUNREACH",
     "ENETRESET",
     "ENETDOWN",
+    // Bun's.
+    "ConnectionRefused",
+    "FailedToOpenSocket",
 ]);
 
-// The subset of the above that can only be reported before a TCP connection
-// exists, and therefore before any request byte could have been written: name
-// resolution produced no address (`ENOTFOUND`, `EAI_AGAIN`) or the peer
-// refused the connection with an RST to the SYN (`ECONNREFUSED`).
+// The subset of the above that can only be reported before any request byte
+// could have been written. Node reports these before a TCP connection exists:
+// name resolution produced no address (`ENOTFOUND`, `EAI_AGAIN`) or the peer
+// refused the connection with an RST to the SYN (`ECONNREFUSED`). Bun gives
+// `ConnectionRefused` for a refused connection, and `ConnectionRefused` or
+// `FailedToOpenSocket` for a name that does not resolve. It also gives
+// `ConnectionRefused` for a TLS handshake that failed, after the TCP
+// connection exists, but the request is only sent once the handshake is done.
 //
 // The routing errnos — `EHOSTUNREACH`, `ENETUNREACH`, `ENETDOWN` — are
 // deliberately absent even though they look like connect-time failures. On
@@ -87,7 +101,13 @@ const TRANSPORT_CODES = new Set([
 // ambiguity this set exists to exclude. They stay in `TRANSPORT_CODES`, so
 // they remain retryable for idempotent calls; only replay eligibility is
 // narrowed. See `isSafeToReplay`.
-const CONNECT_CODES = new Set(["ENOTFOUND", "EAI_AGAIN", "ECONNREFUSED"]);
+const CONNECT_CODES = new Set([
+    "ENOTFOUND",
+    "EAI_AGAIN",
+    "ECONNREFUSED",
+    "ConnectionRefused",
+    "FailedToOpenSocket",
+]);
 
 // Collect every `code` in an error's cause chain. undici does not put the
 // errno on the error it throws — it hangs the underlying socket error off
@@ -105,7 +125,7 @@ const causeCodes = (err: unknown): { codes: string[]; complete: boolean } => {
     return { codes, complete };
 };
 
-// Does the error's cause chain hold the errno of a transport failure, one of
+// Does the error's cause chain hold the code of a transport failure, one of
 // `TRANSPORT_CODES`? `isRetryable` retries such an error, and `classify` in
 // `src/backup.ts` calls it transient.
 export const hasTransportCode = (err: unknown): boolean =>
@@ -152,7 +172,8 @@ export const isRetryable = (err: unknown): boolean => {
     // deliberately literal. Demanding a recognised `cause` instead would
     // classify real network failures as permanent and fail backups that should
     // have succeeded; the cost of the imprecision is bounded by the attempt
-    // count.
+    // count. Bun's fetch rejects with a plain `Error` instead, which is
+    // recognised by its code (see `TRANSPORT_CODES`).
     if (err instanceof TypeError) return true;
     return hasTransportCode(err);
 };
@@ -163,18 +184,18 @@ export const isRetryable = (err: unknown): boolean => {
 // `postJSON` and `putJSON` use this for every `POST` and `PUT` listed in the
 // README under "Endpoints used"; verifying a second factor, for one, consumes
 // one of a small number of attempts. They are replayed only on the failures in
-// `CONNECT_CODES`, which establish that no TCP connection to the server ever
-// existed: there was no address to connect to, or the peer refused the
-// connection outright. A request byte cannot have been transmitted, so the
-// server cannot have acted.
+// `CONNECT_CODES`, which establish that no request byte was ever transmitted:
+// there was no address to connect to, the peer refused the connection
+// outright, or, under bun, the TLS handshake failed. The server cannot have
+// acted.
 //
 // Everything else is ambiguous. A 5xx proves the server did process the
 // request. A reset or a broken pipe can arrive after it was fully sent and
 // acted on. A routing errno can be delivered on an established socket. A
-// deadline says nothing at all about the server's state. So every errno in the
-// cause chain must be a connect errno: one other errno anywhere in the chain
-// is doubt, and doubt is not replayed. A chain longer than the walk is doubt
-// too: the links below the limit were never read.
+// deadline says nothing at all about the server's state. So every code in the
+// cause chain must be one of `CONNECT_CODES`: one other code anywhere in the
+// chain is doubt, and doubt is not replayed. A chain longer than the walk is
+// doubt too: the links below the limit were never read.
 export const isSafeToReplay = (err: unknown): boolean => {
     const { codes, complete } = causeCodes(err);
     return (

test/cli/backup-end-to-end.test.ts

@@ -724,6 +724,50 @@ describe("quak backup", () => {
         },
     );
 
+    // The download of toast.mp4 fails as bun's `fetch` reports a failure, in
+    // the binary `make install` builds: a plain `Error` with bun's own code
+    // and message, and no cause. These are what bun gave for a refused
+    // connection, a name that did not resolve and a reset connection. The
+    // file is tried as often as under Node, and goes into `failures.json`
+    // with that message, as a transient failure.
+    it.each([
+        [
+            "ConnectionRefused",
+            "Unable to connect. Is the computer able to access the url?",
+        ],
+        ["FailedToOpenSocket", "Was there a typo in the url or port?"],
+        [
+            "ECONNRESET",
+            "The socket connection was closed unexpectedly. For more information, pass `verbose: true` in the second argument to fetch()",
+        ],
+    ])(
+        "records a download that failed with bun's %s in failures.json, as transient",
+        async (bunCode, message) => {
+            let attempts = 0;
+            const fetch: typeof globalThis.fetch = async (input, init) => {
+                if (downloadOf(input) === 200) {
+                    attempts++;
+                    throw Object.assign(new Error(message), {
+                        code: bunCode,
+                        errno: 0,
+                    });
+                }
+                return server(input, init);
+            };
+            const dir = mkdtempSync(join(root, "backup-"));
+
+            const { code } = await backup(dir, {}, fetch);
+
+            expect(code).toBe(1);
+            expect(attempts).toBe(UNATTENDED_RETRY_OPTIONS.attempts);
+            const failures = readJSON(join(dir, "failures.json"));
+            expect(failures.files["200"]).toMatchObject({
+                classification: "transient",
+                error: message,
+            });
+        },
+    );
+
     // The name of Ente's API server does not resolve when quak fetches the ML
     // data, after every original has been backed up. No file has ML data
     // cached yet, so each goes into `failures.json` with the reason, as a

test/retry/retry.test.ts

@@ -77,6 +77,14 @@ const recordingSleep = (): {
 const errnoError = (code: string, message = code): Error =>
     Object.assign(new Error(message), { code });
 
+/**
+ * An error shaped like what bun's `fetch` rejects with for a failure below
+ * HTTP: a plain `Error` with bun's own message, a `code`, an `errno` of 0, and
+ * no `cause`.
+ */
+const bunError = (code: string, message: string): Error =>
+    Object.assign(new Error(message), { code, errno: 0 });
+
 // ---------------------------------------------------------------------------
 // Classification
 // ---------------------------------------------------------------------------
@@ -244,6 +252,80 @@ describe("isRetryable: transport failures", () => {
     });
 });
 
+describe("isRetryable and isSafeToReplay: bun's fetch errors", () => {
+    // The binary `make install` builds runs quak under bun, whose `fetch`
+    // does not reject as Node's does. Each shape here was found by running
+    // that binary, built with bun 1.3.13, against listeners on 127.0.0.1 and
+    // a name under `.invalid`.
+
+    it("retries and replays a refused connection", () => {
+        // Bun gives the same error when the TLS handshake fails. The TCP
+        // connection existed then, but the request is only sent once the
+        // handshake is done, so replaying it is as safe.
+        const refused = bunError(
+            "ConnectionRefused",
+            "Unable to connect. Is the computer able to access the url?",
+        );
+        expect(isRetryable(refused)).toBe(true);
+        expect(isSafeToReplay(refused)).toBe(true);
+    });
+
+    it("retries and replays a name that does not resolve, with either code bun gives", () => {
+        // For the same name bun gave `ConnectionRefused` on some requests and
+        // `FailedToOpenSocket` on others.
+        for (const unresolved of [
+            bunError(
+                "ConnectionRefused",
+                "Unable to connect. Is the computer able to access the url?",
+            ),
+            bunError(
+                "FailedToOpenSocket",
+                "Was there a typo in the url or port?",
+            ),
+        ]) {
+            expect(isRetryable(unresolved)).toBe(true);
+            expect(isSafeToReplay(unresolved)).toBe(true);
+        }
+    });
+
+    it("retries a connection closed before the response ended, and does not replay it", () => {
+        // Bun gives Node's errno, `ECONNRESET`, for a reset and for a close,
+        // before the response or part-way through the body. The request may
+        // already have reached the server.
+        const reset = bunError(
+            "ECONNRESET",
+            "The socket connection was closed unexpectedly. For more information, pass `verbose: true` in the second argument to fetch()",
+        );
+        expect(isRetryable(reset)).toBe(true);
+        expect(isSafeToReplay(reset)).toBe(false);
+    });
+
+    it("retries a request that ran out of time, and does not replay it", () => {
+        // What bun's fetch rejects with when `AbortSignal.timeout()`, which
+        // bounds each JSON request, fires.
+        const timedOut = new DOMException(
+            "The operation timed out.",
+            "TimeoutError",
+        );
+        expect(isRetryable(timedOut)).toBe(true);
+        expect(isSafeToReplay(timedOut)).toBe(false);
+    });
+
+    it("does not retry a certificate bun rejects", () => {
+        // A certificate that fails verification fails the same way however
+        // often it is asked. Under Node it is retried only because every
+        // `TypeError` is.
+        expect(
+            isRetryable(
+                bunError(
+                    "DEPTH_ZERO_SELF_SIGNED_CERT",
+                    "self signed certificate",
+                ),
+            ),
+        ).toBe(false);
+    });
+});
+
 describe("isRetryable: stream truncation versus corruption", () => {
     it("retries a truncated stream", () => {
         // Truncation is a transfer that stopped early. The bytes that did