@boxes-dev/dvb 0.2.45 → 0.2.47

package/dist/devbox/commands/init/index.js

@@ -42,6 +42,93 @@ const ensurePrivateDir = async (dir) => {
         // best effort on filesystems that do not support chmod
     }
 };
+const DEFAULT_INIT_STEP_RETRIES = 3;
+const INIT_STEP_RETRYABLE_STATUSES = new Set([
+    408, 409, 425, 429, 500, 502, 503, 504,
+]);
+const delay = async (ms) => new Promise((resolve) => setTimeout(resolve, ms));
+const computeRetryDelayMs = (retryIndex) => {
+    // retryIndex is 1-based (1..N)
+    const base = 500;
+    const cap = 5000;
+    const exp = Math.min(cap, base * 2 ** Math.max(0, retryIndex - 1));
+    const jitter = Math.floor(Math.random() * 150);
+    return exp + jitter;
+};
+const getErrorCode = (error) => {
+    if (!error || typeof error !== "object")
+        return undefined;
+    const record = error;
+    const code = record.code;
+    return typeof code === "string" || typeof code === "number"
+        ? String(code)
+        : undefined;
+};
+const parseExecUnexpectedResponseStatus = (error) => {
+    if (!(error instanceof Error))
+        return null;
+    const match = /Unexpected server response: (\\d{3})/.exec(error.message);
+    if (!match)
+        return null;
+    const status = Number(match[1]);
+    return Number.isFinite(status) ? status : null;
+};
+const isRetryableInitError = (error) => {
+    if (error instanceof SpritesApiError) {
+        return INIT_STEP_RETRYABLE_STATUSES.has(error.status);
+    }
+    // Node's fetch() throws TypeError for network failures.
+    if (error instanceof TypeError)
+        return true;
+    if (error instanceof Error &&
+        (error.name === "AbortError" || error.message.includes("aborted"))) {
+        return true;
+    }
+    const execStatus = parseExecUnexpectedResponseStatus(error);
+    if (execStatus !== null) {
+        return INIT_STEP_RETRYABLE_STATUSES.has(execStatus);
+    }
+    const code = getErrorCode(error);
+    return (code === "ECONNRESET" ||
+        code === "ETIMEDOUT" ||
+        code === "EAI_AGAIN" ||
+        code === "ENOTFOUND" ||
+        code === "ECONNREFUSED");
+};
+const retryInitStep = async ({ status, title, retries = DEFAULT_INIT_STEP_RETRIES, shouldRetry = isRetryableInitError, fn, }) => {
+    let lastError = null;
+    const maxRetries = Math.max(0, Math.floor(retries));
+    const attempts = maxRetries + 1;
+    for (let attempt = 1; attempt <= attempts; attempt += 1) {
+        try {
+            return await fn();
+        }
+        catch (error) {
+            lastError = error;
+            const retryIndex = attempt; // 1-based retry count for the next attempt
+            if (retryIndex > maxRetries || !shouldRetry(error)) {
+                throw error;
+            }
+            status.stage(`${title} (retry ${retryIndex}/${maxRetries})`);
+            logger.warn("init_step_retry", {
+                title,
+                retry: retryIndex,
+                retries: maxRetries,
+                errorName: error instanceof Error ? error.name : undefined,
+                errorMessage: error instanceof Error ? error.message : String(error),
+                errorCode: getErrorCode(error),
+                ...(error instanceof SpritesApiError ? { status: error.status } : {}),
+                ...(parseExecUnexpectedResponseStatus(error) !== null
+                    ? { execStatus: parseExecUnexpectedResponseStatus(error) }
+                    : {}),
+            });
+            await delay(computeRetryDelayMs(retryIndex));
+        }
+    }
+    throw lastError instanceof Error
+        ? lastError
+        : new Error(`Init step failed: ${title}`);
+};
 const migrateLegacyRepoDevboxDir = async ({ repoRoot, projectDir, }) => {
     const legacyDir = path.join(repoRoot, ".devbox");
     try {
@@ -210,14 +297,26 @@ const toRepoRelativePath = (repoRoot, filePath) => {
     return toPosixPath(relative);
 };
 const isOutsideRepoPath = (relativePath) => relativePath === ".." || relativePath.startsWith(`..${path.posix.sep}`);
-const ensureWorkdirOwnership = async ({ client, canonical, workdir, status, json, }) => {
+const ensureWorkdirOwnership = async ({ client, canonical, workdir, status, }) => {
     const checkResult = await client.exec(canonical, [
         "/bin/bash",
         "-lc",
-        `if [ -d ${shellQuote(workdir)} ]; then stat -c %U ${shellQuote(workdir)}; fi`,
+        [
+            "set -euo pipefail",
+            `dir=${shellQuote(workdir)}`,
+            'if [ ! -d "$dir" ]; then',
+            '  echo "Missing workdir: $dir" >&2',
+            "  exit 1",
+            "fi",
+            'stat -c %U "$dir"',
+        ].join("\n"),
     ]);
-    if (checkResult.exitCode !== 0)
-        return;
+    if (checkResult.exitCode !== 0) {
+        const details = checkResult.stderr || checkResult.stdout || "";
+        throw new Error(details
+            ? `Failed to check workdir ownership: ${details.trim()}`
+            : `Failed to check workdir ownership (exit ${checkResult.exitCode})`);
+    }
     const owner = checkResult.stdout.trim();
     if (!owner || owner === "sprite")
         return;
@@ -228,15 +327,10 @@ const ensureWorkdirOwnership = async ({ client, canonical, workdir, status, json
         `sudo -n chown -R sprite:sprite ${shellQuote(workdir)}`,
     ]);
     if (chownResult.exitCode !== 0) {
-        logger.warn("workdir_chown_failed", {
-            box: canonical,
-            workdir,
-            error: chownResult.stderr || chownResult.stdout,
-        });
-        if (!json) {
-            status.stop();
-            console.warn("Warning: failed to update workdir ownership. Git may refuse to operate until ownership is fixed.");
-        }
+        const details = chownResult.stderr || chownResult.stdout || "";
+        throw new Error(details
+            ? `Failed to update workdir ownership: ${details.trim()}`
+            : `Failed to update workdir ownership (exit ${chownResult.exitCode})`);
     }
 };
 const ensureGitSafeDirectory = async ({ client, canonical, workdir, status, }) => {
@@ -573,7 +667,7 @@ export const runInit = async (args) => {
                     fingerprint,
                     phase: label,
                 });
-                return { id: null, comment, createdAt };
+                throw new Error("Checkpoint ID missing.");
             }
             const record = { id, comment, createdAt };
             if (initState) {
@@ -738,13 +832,17 @@ export const runInit = async (args) => {
             });
             await runInitStep({
                 enabled: progressEnabled,
-                title: "Snapshotting filesystem (pre-setup) (optional)",
-                fn: async ({ fail, ok }) => {
+                title: "Snapshotting filesystem (pre-setup)",
+                fn: async ({ status, fail, ok }) => {
                     try {
-                        const checkpoint = await recordCodexCheckpoint({
-                            client,
-                            canonical,
-                            phase: "preCodexSetup",
+                        const checkpoint = await retryInitStep({
+                            status,
+                            title: "Snapshotting filesystem (pre-setup)",
+                            fn: async () => await recordCodexCheckpoint({
+                                client,
+                                canonical,
+                                phase: "preCodexSetup",
+                            }),
                         });
                         if (checkpoint.id) {
                             ok(`Snapshot created: ${checkpoint.id}`);
@@ -757,10 +855,8 @@ export const runInit = async (args) => {
                             phase: "pre-codex-setup",
                             error: error instanceof Error ? error.message : String(error),
                         });
-                        fail("Snapshotting filesystem (pre-setup) (optional) (failed)");
-                        if (!parsed.json) {
-                            console.warn("Warning: failed to create pre-setup filesystem snapshot. Continuing without it.");
-                        }
+                        fail("Snapshotting filesystem (pre-setup) (failed)");
+                        throw error;
                     }
                 },
             });
@@ -796,13 +892,17 @@ export const runInit = async (args) => {
             });
             await runInitStep({
                 enabled: progressEnabled,
-                title: "Snapshotting filesystem (post-setup) (optional)",
-                fn: async ({ fail, ok }) => {
+                title: "Snapshotting filesystem (post-setup)",
+                fn: async ({ status, fail, ok }) => {
                     try {
-                        const checkpoint = await recordCodexCheckpoint({
-                            client,
-                            canonical,
-                            phase: "postCodexSetup",
+                        const checkpoint = await retryInitStep({
+                            status,
+                            title: "Snapshotting filesystem (post-setup)",
+                            fn: async () => await recordCodexCheckpoint({
+                                client,
+                                canonical,
+                                phase: "postCodexSetup",
+                            }),
                         });
                         if (checkpoint.id) {
                             ok(`Snapshot created: ${checkpoint.id}`);
@@ -815,10 +915,8 @@ export const runInit = async (args) => {
                             phase: "post-codex-setup",
                             error: error instanceof Error ? error.message : String(error),
                         });
-                        fail("Snapshotting filesystem (post-setup) (optional) (failed)");
-                        if (!parsed.json) {
-                            console.warn("Warning: failed to create post-setup filesystem snapshot. Continuing without it.");
-                        }
+                        fail("Snapshotting filesystem (post-setup) (failed)");
+                        throw error;
                     }
                 },
             });
@@ -956,25 +1054,20 @@ export const runInit = async (args) => {
             localPaths: projectLocalPaths,
         });
         const updateRegistryProjectStatus = async (initStatus) => {
-            try {
-                await requestJson(socketInfo.socketPath, "POST", "/registry/upsert", DAEMON_TIMEOUT_MS.registry, {
-                    project: buildProjectEntry(initStatus),
-                });
-            }
-            catch (error) {
-                logger.warn("registry_project_update_failed", {
-                    box: canonical,
-                    status: initStatus,
-                    error: String(error),
-                });
-            }
+            await requestJson(socketInfo.socketPath, "POST", "/registry/upsert", DAEMON_TIMEOUT_MS.registry, {
+                project: buildProjectEntry(initStatus),
+            });
         };
         await runInitStep({
             enabled: progressEnabled,
             title: "Bootstrapping devbox",
-            fn: async ({ fail }) => {
+            fn: async ({ status, fail }) => {
                 try {
-                    await bootstrapDevbox(client, canonical);
+                    await retryInitStep({
+                        status,
+                        title: "Bootstrapping devbox",
+                        fn: async () => await bootstrapDevbox(client, canonical),
+                    });
                 }
                 catch (error) {
                     logger.warn("devbox_bootstrap_failed", {
@@ -982,12 +1075,7 @@ export const runInit = async (args) => {
                         error: String(error),
                     });
                     fail("Bootstrapping devbox (failed)");
-                    if (!parsed.json) {
-                        const message = error instanceof Error && error.message
-                            ? error.message
-                            : String(error);
-                        console.warn(`Warning: devbox bootstrap failed. ${message}`);
-                    }
+                    throw error;
                 }
             },
         });
@@ -1019,7 +1107,7 @@ export const runInit = async (args) => {
             await runInitStep({
                 enabled: progressEnabled,
                 title: "Installing sprite daemon",
-                fn: async ({ fail }) => {
+                fn: async ({ status, fail }) => {
                     try {
                         const convexUrl = getConvexUrl();
                         if (!controlPlaneToken) {
@@ -1036,12 +1124,16 @@ export const runInit = async (args) => {
                         if (!heartbeatToken) {
                             throw new Error("Daemon token unavailable.");
                         }
-                        await installSpriteDaemon({
-                            client,
-                            spriteName: canonical,
-                            release,
-                            convexUrl,
-                            heartbeatToken,
+                        await retryInitStep({
+                            status,
+                            title: "Installing sprite daemon",
+                            fn: async () => await installSpriteDaemon({
+                                client,
+                                spriteName: canonical,
+                                release,
+                                convexUrl,
+                                heartbeatToken,
+                            }),
                         });
                         await updateInitState({ steps: { daemonInstalled: true } });
                     }
@@ -1051,12 +1143,7 @@ export const runInit = async (args) => {
                             error: String(error),
                         });
                         fail("Installing sprite daemon (failed)");
-                        if (!parsed.json) {
-                            const message = error instanceof Error && error.message
-                                ? error.message
-                                : String(error);
-                            console.warn(`Warning: failed to install sprite daemon. ${message}`);
-                        }
+                        throw error;
                     }
                 },
             });
@@ -1064,9 +1151,13 @@ export const runInit = async (args) => {
         await runInitStep({
             enabled: progressEnabled,
             title: "Ensuring sprite daemon service",
-            fn: async ({ fail }) => {
+            fn: async ({ status, fail }) => {
                 try {
-                    await ensureSpriteDaemonService({ client, spriteName: canonical });
+                    await retryInitStep({
+                        status,
+                        title: "Ensuring sprite daemon service",
+                        fn: async () => await ensureSpriteDaemonService({ client, spriteName: canonical }),
+                    });
                     await updateInitState({ steps: { daemonServiceEnsured: true } });
                 }
                 catch (error) {
@@ -1075,12 +1166,7 @@ export const runInit = async (args) => {
                         error: String(error),
                     });
                     fail("Ensuring sprite daemon service (failed)");
-                    if (!parsed.json) {
-                        const message = error instanceof Error && error.message
-                            ? error.message
-                            : String(error);
-                        console.warn(`Warning: failed to ensure sprite daemon service. ${message}`);
-                    }
+                    throw error;
                 }
             },
         });
@@ -1897,12 +1983,15 @@ export const runInit = async (args) => {
             enabled: progressEnabled,
             title: "Ensuring workdir ownership",
             fn: async ({ status }) => {
-                await ensureWorkdirOwnership({
-                    client,
-                    canonical,
-                    workdir: expandedWorkdir,
+                await retryInitStep({
                     status,
-                    json: parsed.json === true,
+                    title: "Ensuring workdir ownership",
+                    fn: async () => await ensureWorkdirOwnership({
+                        client,
+                        canonical,
+                        workdir: expandedWorkdir,
+                        status,
+                    }),
                 });
             },
         });
@@ -1927,13 +2016,19 @@ export const runInit = async (args) => {
             await runInitStep({
                 enabled: progressEnabled,
                 title: "Configuring SSH mount access",
-                fn: async ({ fail }) => {
+                fn: async ({ status, fail }) => {
                     try {
-                        await ensureLocalMountKey();
-                        await ensureKnownHostsFile();
-                        const publicKey = await readLocalMountPublicKey();
-                        await ensureRemoteMountAccess(client, canonical, publicKey);
-                        await ensureSshdService(client, canonical);
+                        await retryInitStep({
+                            status,
+                            title: "Configuring SSH mount access",
+                            fn: async () => {
+                                await ensureLocalMountKey();
+                                await ensureKnownHostsFile();
+                                const publicKey = await readLocalMountPublicKey();
+                                await ensureRemoteMountAccess(client, canonical, publicKey);
+                                await ensureSshdService(client, canonical);
+                            },
+                        });
                         await updateInitState({ steps: { sshdConfigured: true } });
                     }
                     catch (error) {
@@ -1942,13 +2037,7 @@ export const runInit = async (args) => {
                             error: String(error),
                         });
                         fail("Configuring SSH mount access (failed)");
-                        if (!parsed.json) {
-                            const message = error instanceof Error && error.message
-                                ? error.message
-                                : String(error);
-                            console.warn(`Warning: failed to configure SSH mount access. ${message}`);
-                            console.warn(`To retry later: dvb mount ${alias}`);
-                        }
+                        throw error;
                     }
                 },
             });
@@ -2086,10 +2175,12 @@ export const runInit = async (args) => {
         }
         const weztermMuxPresent = await runInitStep({
             enabled: progressEnabled,
-            title: "Ensuring WezTerm mux server (optional)",
-            fn: async ({ fail }) => {
-                try {
-                    const installResult = !shouldResume || !initState?.steps.weztermMuxInstalled
+            title: "Ensuring WezTerm mux server",
+            fn: async ({ status }) => {
+                const installResult = await retryInitStep({
+                    status,
+                    title: "Ensuring WezTerm mux server",
+                    fn: async () => !shouldResume || !initState?.steps.weztermMuxInstalled
                         ? await ensureWeztermMuxInstalled({
                             client,
                             spriteName: canonical,
@@ -2100,52 +2191,29 @@ export const runInit = async (args) => {
                             spriteName: canonical,
                             // If the mux is missing on resume, we still need to resolve a GitHub asset.
                             allowResolveAsset: true,
-                        });
-                    if (installResult.muxPresent) {
-                        await updateInitState({ steps: { weztermMuxInstalled: true } });
-                    }
-                    return installResult.muxPresent;
-                }
-                catch (error) {
-                    logger.warn("wezterm_mux_install_failed", {
-                        box: canonical,
-                        error: error instanceof Error ? error.message : String(error),
-                    });
-                    fail("Ensuring WezTerm mux server (optional) (failed)");
-                    if (!parsed.json) {
-                        const message = error instanceof Error && error.message
-                            ? error.message
-                            : String(error);
-                        console.warn(`Warning: failed to install WezTerm mux server (optional). ${message}\n` +
-                            "Tip: re-run with DEVBOX_LOG_LEVEL=info to see Sprite exec logs on stderr.");
-                    }
-                    return false;
+                        }),
+                });
+                if (installResult.muxPresent) {
+                    await updateInitState({ steps: { weztermMuxInstalled: true } });
+                    return true;
                 }
+                throw new Error("WezTerm mux server unavailable.");
             },
         });
-        if (weztermMuxPresent &&
-            (!shouldResume || !initState?.steps.weztermMuxServiceEnsured)) {
+        if (!shouldResume || !initState?.steps.weztermMuxServiceEnsured) {
             await runInitStep({
                 enabled: progressEnabled,
-                title: "Ensuring WezTerm mux service (optional)",
-                fn: async ({ fail }) => {
-                    try {
-                        await ensureWeztermMuxService({ client, spriteName: canonical });
-                        await updateInitState({ steps: { weztermMuxServiceEnsured: true } });
-                    }
-                    catch (error) {
-                        logger.warn("wezterm_mux_service_failed", {
-                            box: canonical,
-                            error: error instanceof Error ? error.message : String(error),
-                        });
-                        fail("Ensuring WezTerm mux service (optional) (failed)");
-                        if (!parsed.json) {
-                            const message = error instanceof Error && error.message
-                                ? error.message
-                                : String(error);
-                            console.warn(`Warning: failed to ensure WezTerm mux service (optional). ${message}`);
-                        }
+                title: "Ensuring WezTerm mux service",
+                fn: async ({ status }) => {
+                    if (!weztermMuxPresent) {
+                        throw new Error("WezTerm mux server unavailable.");
                     }
+                    await retryInitStep({
+                        status,
+                        title: "Ensuring WezTerm mux service",
+                        fn: async () => await ensureWeztermMuxService({ client, spriteName: canonical }),
+                    });
+                    await updateInitState({ steps: { weztermMuxServiceEnsured: true } });
                 },
             });
         }
@@ -2213,20 +2281,13 @@ export const runInit = async (args) => {
         });
         await runInitStep({
             enabled: progressEnabled,
-            title: "Updating Codex config on sprite (optional)",
-            fn: async ({ fail }) => {
-                try {
-                    await writeRemoteCodexConfig(client, canonical, repoName);
-                }
-                catch (error) {
-                    logger.warn("codex_config_update_failed", {
-                        error: String(error),
-                    });
-                    fail("Updating Codex config on sprite (optional) (failed)");
-                    if (!parsed.json) {
-                        console.warn("Warning: failed to update /home/sprite/.codex/config.toml for full-access sandbox settings.");
-                    }
-                }
+            title: "Updating Codex config on sprite",
+            fn: async ({ status }) => {
+                await retryInitStep({
+                    status,
+                    title: "Updating Codex config on sprite",
+                    fn: async () => await writeRemoteCodexConfig(client, canonical, repoName),
+                });
             },
         });
         const skipSetupArtifactsStage = skipCodexApply ||
@@ -2251,13 +2312,17 @@ export const runInit = async (args) => {
         if (!skipCodexApply) {
             await runInitStep({
                 enabled: progressEnabled,
-                title: "Snapshotting filesystem (pre-setup) (optional)",
-                fn: async ({ fail, ok }) => {
+                title: "Snapshotting filesystem (pre-setup)",
+                fn: async ({ status, fail, ok }) => {
                     try {
-                        const checkpoint = await recordCodexCheckpoint({
-                            client,
-                            canonical,
-                            phase: "preCodexSetup",
+                        const checkpoint = await retryInitStep({
+                            status,
+                            title: "Snapshotting filesystem (pre-setup)",
+                            fn: async () => await recordCodexCheckpoint({
+                                client,
+                                canonical,
+                                phase: "preCodexSetup",
+                            }),
                         });
                         if (checkpoint.id) {
                             ok(`Snapshot created: ${checkpoint.id}`);
@@ -2270,10 +2335,8 @@ export const runInit = async (args) => {
                             phase: "pre-codex-setup",
                             error: error instanceof Error ? error.message : String(error),
                         });
-                        fail("Snapshotting filesystem (pre-setup) (optional) (failed)");
-                        if (!parsed.json) {
-                            console.warn("Warning: failed to create pre-setup filesystem snapshot. Continuing without it.");
-                        }
+                        fail("Snapshotting filesystem (pre-setup) (failed)");
+                        throw error;
                     }
                 },
             });
@@ -2314,13 +2377,17 @@ export const runInit = async (args) => {
             await updateRegistryProjectStatus(resolveInitStatus(initState?.steps, initState?.complete));
             await runInitStep({
                 enabled: progressEnabled,
-                title: "Snapshotting filesystem (post-setup) (optional)",
-                fn: async ({ fail, ok }) => {
+                title: "Snapshotting filesystem (post-setup)",
+                fn: async ({ status, fail, ok }) => {
                     try {
-                        const checkpoint = await recordCodexCheckpoint({
-                            client,
-                            canonical,
-                            phase: "postCodexSetup",
+                        const checkpoint = await retryInitStep({
+                            status,
+                            title: "Snapshotting filesystem (post-setup)",
+                            fn: async () => await recordCodexCheckpoint({
+                                client,
+                                canonical,
+                                phase: "postCodexSetup",
+                            }),
                         });
                         if (checkpoint.id) {
                             ok(`Snapshot created: ${checkpoint.id}`);
@@ -2333,10 +2400,8 @@ export const runInit = async (args) => {
                             phase: "post-codex-setup",
                             error: error instanceof Error ? error.message : String(error),
                         });
-                        fail("Snapshotting filesystem (post-setup) (optional) (failed)");
-                        if (!parsed.json) {
-                            console.warn("Warning: failed to create post-setup filesystem snapshot. Continuing without it.");
-                        }
+                        fail("Snapshotting filesystem (post-setup) (failed)");
+                        throw error;
                     }
                 },
             });