fix(github): report a failed apply's furthest copy progress and stop claiming unverified checkpoint resumes

[aparajon]

Two operator-facing accuracy fixes for applies that went through operator retries:

• "Failed at N%" now reports the run's furthest copy progress. Each retry relaunches the engine's row copy, so the live per-table progress resets with every attempt — the failure label was showing whichever attempt was sampled last, which can understate a run whose earlier attempt got further (or read as a completed copy when an earlier attempt hit 100%). The tasks table gains best_rows_copied / best_progress_percent high-water columns (both dialects), raised by the store on every progress write and never allowed to regress; the terminal summary label and the in-progress failed row's bar render from the high-water. A storage parity test pins the never-regress contract on MySQL and PostgreSQL.

attempt 1 copy ──────────────▶ 47% ✗ interrupted
attempt 2 copy ────▶ 12% ✗ failed (sampled last)

live progress:   12%   ── what the comment used to say
high-water:      47%   ── what it says now

• The driver no longer claims "resumed from checkpoint" before the engine confirms it. The resume timeline event named the code path, not the outcome — it claimed a checkpoint resume even when the engine found no checkpoint and restarted the copy from scratch. The relaunch event now says only that the apply resumed; the checkpoint claim comes solely from the engine-confirmed resume event, and the engine's own "could not resume from checkpoint" log line covers the restart case. The deferred-cutover recovery event now names what was actually verified (the engine cutover signal).

Summary: Failed (Retries Exhausted) — the failure label reports the run's high-water (47%), not the last-sampled attempt (12%)

❌ Schema Change Failed — Staging

Database: testapp | Apply ID: apply-a1b2c3d4e5f6 | Duration: 2h 15m

Applied by @jackjackbits at 2026-03-15 12:15:00 UTC

⚠️ Error: lock wait timeout exceeded; try restarting transaction

1 of 2 tables completed before failure.

users — Failed at 47%

ALTER TABLE `users` ADD INDEX `idx_email`(`email`);

orders

ALTER TABLE `orders` ADD INDEX `idx_user_id`(`user_id`);

---

To retry:

schemabot apply -e staging

Note

Deployed PostgreSQL storage databases need the two new tasks columns added by hand before this deploys (PostgreSQL bootstrap verifies columns but does not alter existing tables).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR updates SchemaBot’s GitHub-facing rendering and timeline events so operator-visible surfaces don’t claim unverified checkpoint resumes and don’t report misleading row-copy progress when an apply has been retried (where the stored per-table percent can reflect an arbitrary attempt).

Changes:

• Omit progress bars / “Failed at N%” labels on failed rows and summary entries when Attempt > 0 (the apply has been redispatched at least once), while keeping the percentage for single-attempt failures.
• Adjust LocalClient resume/recovery timeline messages to avoid claiming “resumed from checkpoint” unless the engine confirms it, and add coverage to enforce that invariant.
• Add a new preview type + template snippet for “Summary: Failed (Retries Exhausted)” and corresponding rendering/tests.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
TEMPLATES.md	Adds a rendered template example for “Summary: Failed (Retries Exhausted)”.
pkg/webhook/templates/preview.go	Adds a preview renderer for the retry-exhausted failed summary.
pkg/webhook/templates/apply.go	Suppresses failed-row progress bars and “Failed at N%” in multi-attempt runs; threads multi-attempt flag into summary table entries.
pkg/webhook/templates/apply_test.go	Adds tests ensuring multi-attempt failures omit the bar/percent while single-attempt failures retain them.
pkg/tern/local_control_resume.go	Updates relaunch/recovery timeline event messages to avoid unverified “checkpoint” claims.
pkg/tern/local_control_resume_test.go	Adds a test asserting relaunch events do not contain “checkpoint” unless engine-confirmed.
pkg/cmd/internal/templates/preview.go	Registers a new preview type for “failed after retries exhausted”.
pkg/cmd/internal/templates/preview_dispatch.go	Wires the new preview type to the webhook template preview renderer.
pkg/cmd/internal/templates/preview_comment.go	Includes the new summary preview in the “all previews” outputs.
pkg/cmd/commands/preview.go	Adds the new preview type to the CLI’s allowed preview set.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.