refactor(release-sandbox): remove redundant commit message check from workflow

Updated the GitHub Actions workflow for the studio-sandbox release by removing the conditional check that skipped republishing for commits with a '[release]:' prefix. This simplifies the workflow logic.

feat(sandbox): implement retry logic for transient network errors during cloning

Enhanced the cloning process in the sandbox daemon by adding retry logic for transient network errors. The new implementation allows up to three retries with a delay between attempts, improving robustness against temporary network issues.

Author: pedrofrxncx

.github/workflows/release-studio-sandbox.yaml

@@ -15,10 +15,6 @@ jobs:
   build-push:
     name: Build & Push studio-sandbox image
     runs-on: ubuntu-latest
-    # `[release]:` commits come from the auto-bump bot and don't actually
-    # change the sandbox source — skip them to avoid republishing the same
-    # version on every release tag bump.
-    if: ${{ !startsWith(github.event.head_commit.message, '[release]:') }}
     permissions:
       contents: read
       packages: write

packages/sandbox/daemon/setup/clone.ts

@@ -29,6 +29,48 @@ function runStep(cmd: string, deps: CloneDeps): Promise<number> {
   return spawnSetupStep(cmd, deps.onChunk, deps.dropPrivileges);
 }
 
+const TRANSIENT_ERRORS = [
+  "Could not resolve host",
+  "early EOF",
+  "unexpected disconnect",
+  "Connection reset by peer",
+  "Connection timed out",
+];
+const CLONE_MAX_RETRIES = 3;
+const CLONE_RETRY_DELAY_MS = 3000;
+
+function isTransient(output: string): boolean {
+  return TRANSIENT_ERRORS.some((e) => output.includes(e));
+}
+
+function sleep(ms: number): Promise<void> {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
+
+async function runNetworkStep(cmd: string, deps: CloneDeps): Promise<number> {
+  for (let attempt = 0; attempt <= CLONE_MAX_RETRIES; attempt++) {
+    if (attempt > 0) {
+      deps.onChunk(
+        "setup",
+        `\r\n[clone] transient network error, retrying in ${CLONE_RETRY_DELAY_MS / 1000}s (attempt ${attempt + 1}/${CLONE_MAX_RETRIES + 1})...\r\n`,
+      );
+      await sleep(CLONE_RETRY_DELAY_MS);
+    }
+    let output = "";
+    const tee: CloneDeps = {
+      ...deps,
+      onChunk: (src, data) => {
+        output += data;
+        deps.onChunk(src, data);
+      },
+    };
+    const code = await runStep(cmd, tee);
+    if (code === 0) return 0;
+    if (!isTransient(output) || attempt >= CLONE_MAX_RETRIES) return code;
+  }
+  return 1;
+}
+
 /** Resolves to exit code (0 on success). Emits chunks via `onChunk`. */
 export async function spawnClone(deps: CloneDeps): Promise<number> {
   const { config } = deps;
@@ -52,18 +94,21 @@ export async function spawnClone(deps: CloneDeps): Promise<number> {
   // and is not an empty directory". Use init+fetch+checkout instead — it
   // operates in-place and tolerates existing content.
   if (isNonEmptyWithoutGit(dir)) {
-    const steps = [
+    const localSteps = [
       `${gc} -C ${dir} init`,
       `${gc} -C ${dir} remote add origin ${cloneUrl}`,
-      `${gc} -C ${dir} fetch --depth 1 origin HEAD`,
-      `${gc} -C ${dir} checkout FETCH_HEAD`,
     ];
-    for (const step of steps) {
+    for (const step of localSteps) {
       const code = await runStep(step, deps);
       if (code !== 0) return code;
     }
-    return 0;
+    const fetchCode = await runNetworkStep(
+      `${gc} -C ${dir} fetch --depth 1 origin HEAD`,
+      deps,
+    );
+    if (fetchCode !== 0) return fetchCode;
+    return runStep(`${gc} -C ${dir} checkout FETCH_HEAD`, deps);
   }
 
-  return runStep(`${gc} clone --depth 1 ${cloneUrl} ${dir}`, deps);
+  return runNetworkStep(`${gc} clone --depth 1 ${cloneUrl} ${dir}`, deps);
 }