@bli-cockpit/cli 0.2.101 → 0.2.102

package/dist/commands/doctor-pipeline.js

@@ -14,7 +14,7 @@ import { runBackfillCommand } from "./backfill.js";
 import { diskRowMessage, mib, redeliveryLine } from "./doctor-disk-words.js";
 import { doctorRoots } from "./doctor-access.js";
 import { backfillCompletionStepState, backfillFixVerdict, jsonField, parseDoctorBackfillJson, parseDoctorSyncJson, syncBacklogDrainingVerdict, syncStandAsideVerdict, } from "./doctor-pipeline-verdicts.js";
-import { fail, needsFix, ok, skipped } from "./doctor-report.js";
+import { asRecord, fail, needsFix, ok, skipped } from "./doctor-report.js";
 /**
  * The `backfill-complete`, `gc-checked`, `disk-bounded`, and `sync-fresh`
  * check family: does the collection pipeline itself have everything it
@@ -47,33 +47,69 @@ export async function checkBackfillState(context) {
     return needsFix("backfill-complete", cursor.updated_at ? "partial" : "never_run", "the catch-up over your old sessions has not finished");
 }
 export async function fixBackfillState(context) {
-    return withDoctorLockWait(context, () => runBackfillRepair(context));
-}
-async function runBackfillRepair(context) {
-    const capture = capturedIo(context.io, !context.command.json);
-    const code = await runBackfillCommand({
-        homeDir: context.command.homeDir,
-        repoRoot: context.command.repoRoot,
-        all: true,
-        dryRun: false,
-        yes: true,
-        json: true,
-    }, capture.io);
-    const stdout = capture.stdout();
-    const output = `${stdout}\n${capture.stderr()}`;
-    if (code === 0) {
-        // Re-read the marker this run just wrote instead of hand-rolling a second
-        // message: `checkBackfillState`'s pure core already knows how to say
-        // "complete" vs "complete_with_oversized_skips" (BLI-2727), and this way
-        // the two can never say something different for the same marker.
-        const recheck = await checkBackfillState(context);
-        if (recheck.status === "ok")
-            return recheck;
-        return recheck;
+    const started = Date.now();
+    const deadline = started + (context.command.backfillBudgetSeconds ?? 900) * 1000;
+    let caughtUp = 0;
+    let remaining = null;
+    const unfinished = (code, reason) => ({
+        ...needsFix("backfill-complete", code, `caught up on ${caughtUp} old sessions this run, ${remaining ?? "unknown"} still to go; ${reason}; run \`cockpit doctor\` again to continue`),
+        nextAction: "cockpit doctor",
+    });
+    const progress = setInterval(() => {
+        const seconds = Math.floor((Date.now() - started) / 1000);
+        context.io.stderr.write(`backfill: ${caughtUp} caught up, ${remaining ?? "unknown"} to go, ${Math.floor(seconds / 60)}m${seconds % 60}s\n`);
+        console.error("[doctor] backfill-complete progress", JSON.stringify({ caught_up: caughtUp, remaining, elapsed_seconds: seconds }));
+    }, 30_000);
+    try {
+        while (Date.now() < deadline) {
+            // Every chunk reacquires the collector's locks. Bound lock waiting by
+            // the overall budget too; never abandon an in-flight cursor write.
+            const chunkContext = { ...context, command: { ...context.command,
+                    lockWaitSeconds: Math.min(context.command.lockWaitSeconds ?? 600, (deadline - Date.now()) / 1000),
+                } };
+            const row = await withDoctorLockWait(chunkContext, async () => {
+                if (Date.now() >= deadline)
+                    return unfinished("backfill_budget_timeout", "backfill time budget reached");
+                const capture = capturedIo(context.io, false);
+                const code = await runBackfillCommand({
+                    homeDir: context.command.homeDir, repoRoot: context.command.repoRoot,
+                    all: true, dryRun: false, yes: true, json: true,
+                }, capture.io);
+                const parsed = parseDoctorBackfillJson(capture.stdout());
+                const counts = asRecord(parsed?.counts);
+                const count = (value) => typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : null;
+                caughtUp += count(counts?.["backfilled"]) ?? 0;
+                remaining = count(counts?.["remaining"]) ?? remaining;
+                if (code === 0 && parsed?.status === "complete") {
+                    // Reporting totals include permanent caps. Preserve their marker
+                    // note instead of retrying history the collector says is complete.
+                    const checked = await checkBackfillState(context);
+                    return ok("backfill-complete", checked.status === "ok" ? checked.code : "complete", `caught up on ${caughtUp} old sessions this run` + (checked.status === "ok" ? `; ${checked.message}` : ""));
+                }
+                const verdict = backfillFixVerdict(parsed, jsonField(capture.stderr(), "failure_reason"));
+                if (parsed?.status === "partial" &&
+                    (parsed.failure_reason === "deferred_budget_exhausted" || !parsed.failure_reason)) {
+                    return needsFix("backfill-complete", "backfill_chunk_pending", "more history remains");
+                }
+                return { ...verdict, message: typeof parsed?.failure_reason === "string" ? parsed.failure_reason : "backfill failed without a valid completion receipt" };
+            });
+            if (row.status === "ok") {
+                console.error("[doctor] backfill-complete finished", JSON.stringify({ caught_up: caughtUp, remaining, elapsed_seconds: Math.floor((Date.now() - started) / 1000) }));
+                return row;
+            }
+            if (row.code !== "backfill_chunk_pending") {
+                return unfinished(row.code === "lock_wait_timeout" && Date.now() >= deadline ? "backfill_budget_timeout" : row.code, row.message);
+            }
+        }
+        return unfinished("backfill_budget_timeout", "backfill time budget reached");
+    }
+    catch (error) {
+        console.error("[doctor] backfill-complete failed", JSON.stringify({ reason: "backfill_threw", error_name: error instanceof Error ? error.name : "unknown" }));
+        return unfinished("backfill_threw", "backfill failed");
+    }
+    finally {
+        clearInterval(progress);
     }
-    const verdict = backfillFixVerdict(parseDoctorBackfillJson(stdout), jsonField(output, "failure_reason"));
-    console.error("[cockpit-doctor] catch-up run did not finish", JSON.stringify({ reason: verdict.code, row_status: verdict.status, exit_code: code }));
-    return verdict;
 }
 export async function checkGcState(context) {
     if (context.io.env["COCKPIT_DISABLE_GC"] === "1") {

package/dist/commands/doctor-update.js

@@ -71,6 +71,8 @@ export function reexecDoctor(command, io) {
     const args = ["doctor"];
     if (command.homeDir)
         args.push("--home", command.homeDir);
+    if (command.backfillBudgetSeconds)
+        args.push("--backfill-budget", String(command.backfillBudgetSeconds));
     if (command.lockWaitSeconds)
         args.push("--lock-wait", String(command.lockWaitSeconds));
     if (command.allowHomeRoot)

package/dist/commands/local-args-collector-setup.js

@@ -101,6 +101,7 @@ export function parseDoctorArgs(alias, args) {
             "--check",
             "--no-repair",
             "--lock-wait",
+            "--backfill-budget",
             "--json",
             "--allow-home-root",
             "--max-depth",
@@ -117,6 +118,7 @@ export function parseDoctorArgs(alias, args) {
         ],
         valueFlags: [
             "--lock-wait",
+            "--backfill-budget",
             "--home",
             "--repo",
             "--workspace",
@@ -140,6 +142,7 @@ export function parseDoctorArgs(alias, args) {
     return {
         kind: "doctor",
         checkOnly: values.booleans.has("--check") || values.booleans.has("--no-repair"),
+        backfillBudgetSeconds: optionalPositiveInteger(values.flags.get("--backfill-budget"), "--backfill-budget") ?? 900,
         lockWaitSeconds: optionalPositiveInteger(values.flags.get("--lock-wait"), "--lock-wait") ?? 600,
         alias,
         homeDir: optionalNonEmpty(values.flags.get("--home")),

package/dist/commands/local-help-commands.js

@@ -71,11 +71,12 @@ export function localSubcommandHelp(command) {
         [
             "do-everything",
             [
-                "Usage: cockpit do-everything [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--json]",
+                "Usage: cockpit do-everything [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--backfill-budget <seconds>] [--json]",
                 "",
                 "Diagnose, repair, and verify this machine. doctor, do-everything, and fix run the same job.",
                 "Repairs CLI, background sync, sign-in, roots, memory and hooks, agent rules, catch-up, uploads, and cleanup.",
                 "--check or --no-repair diagnoses without repairs. --dry-run previews repairs.",
+                "--backfill-budget defaults to 900 seconds; catch-up repeats chunks until complete, with progress every 30 seconds.",
                 "--lock-wait defaults to 600 seconds; progress prints every 30 seconds.",
                 "--json includes steps, repairs, and needs_person. Exit 0 means nothing needs you.",
                 "Maintainers: --update-tag next keeps self-update on the prerelease candidate.",
@@ -84,11 +85,12 @@ export function localSubcommandHelp(command) {
         [
             "fix",
             [
-                "Usage: cockpit fix [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--json]",
+                "Usage: cockpit fix [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--backfill-budget <seconds>] [--json]",
                 "",
                 "Diagnose, repair, and verify this machine. doctor, do-everything, and fix run the same job.",
                 "Repairs CLI, background sync, sign-in, roots, memory and hooks, agent rules, catch-up, uploads, and cleanup.",
                 "--check or --no-repair diagnoses without repairs. --dry-run previews repairs.",
+                "--backfill-budget defaults to 900 seconds; catch-up repeats chunks until complete, with progress every 30 seconds.",
                 "--lock-wait defaults to 600 seconds; progress prints every 30 seconds.",
                 "--json includes steps, repairs, and needs_person. Exit 0 means nothing needs you.",
                 "Maintainers: --update-tag next keeps self-update on the prerelease candidate.",
@@ -97,11 +99,12 @@ export function localSubcommandHelp(command) {
         [
             "doctor",
             [
-                "Usage: cockpit doctor [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--json]",
+                "Usage: cockpit doctor [--workspace <path>] [--check | --no-repair] [--lock-wait <seconds>] [--backfill-budget <seconds>] [--json]",
                 "",
                 "Diagnose, repair, and verify this machine. doctor, do-everything, and fix run the same job.",
                 "Repairs CLI, background sync, sign-in, roots, memory and hooks, agent rules, catch-up, uploads, and cleanup.",
                 "--check or --no-repair diagnoses without repairs. --dry-run previews repairs.",
+                "--backfill-budget defaults to 900 seconds; catch-up repeats chunks until complete, with progress every 30 seconds.",
                 "--lock-wait defaults to 600 seconds; progress prints every 30 seconds.",
                 "--json includes steps, repairs, and needs_person. Exit 0 means nothing needs you.",
                 "Maintainers: --update-tag next keeps self-update on the prerelease candidate.",

package/dist/commands/local-help.js

@@ -64,7 +64,7 @@ export function localCommandHelp(command) {
         "  cockpit upgrade [same flags as update]",
         "  cockpit do-everything [--workspace <path>] [--dashboard-url <url>] [--update-tag <tag>] [--dry-run] [--json]",
         "  cockpit fix [same flags as do-everything]",
-        "  cockpit doctor [--check | --no-repair] [--lock-wait <seconds>] [--json]",
+        "  cockpit doctor [--check | --no-repair] [--lock-wait <seconds>] [--backfill-budget <seconds>] [--json]",
         "  cockpit install [--dashboard-url <url>] [--workspace <path>] [--allow-home-root] [--json]",
         "  cockpit login [--pair <code>] [--no-browser] [--legacy-pair] [--email <owner@email>] [--device-name <name>] [--dashboard-url <url>] [--no-auth] [--json]",
         "  cockpit pair [--pair <code>] [--no-browser] [--legacy-pair] [--email <owner@email>] [--device-name <name>] [--dashboard-url <url>] [--no-auth] [--json]",

package/dist/commands/public-root.js

@@ -15,7 +15,7 @@ export async function runCockpitCli(argv, io) {
   }
 
   if (command === "--version" || command === "-V" || command === "version") {
-    writeLine(io?.stdout ?? process.stdout, "0.2.101");
+    writeLine(io?.stdout ?? process.stdout, "0.2.102");
     return 0;
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bli-cockpit/cli",
-  "version": "0.2.101",
+  "version": "0.2.102",
   "private": false,
   "type": "module",
   "bin": {