Pre-launch — Gluecron is in final validation. Public signups and git hosting for non-owner users open after launch review.
CodeIssuesDiscussionsWikiPull RequestsProjectsCommitsActionsReleasesContributorsPulse● GatesSecuritySettingsDeploymentsPipelineInsightsAgents✨ Explain✨ Ask AI✨ Workspace✨ Spec✨ Tests▓ Debt Map✨ NL Search🏛 Archaeology
claude/adoring-hopper-5x74bqclaude/affectionate-feynman-ykrf1hclaude/architecture-audit-design-wxprenclaude/build-status-update-3MXsfclaude/charming-meitner-mllb5rclaude/compare-gate-gluecron-s4mFQclaude/confident-faraday-tikcwbclaude/continue-work-XMTlIclaude/crontech-gluecron-deploy-7MIECclaude/crontech-platform-setup-SeKfwclaude/design-2026claude/ecstatic-ptolemy-jMdigclaude/enhance-github-integration-QNHdGclaude/fix-aa-loop-issue-PonMQclaude/fix-actions-and-processclaude/fix-desktop-errors-XqoW8claude/fix-red-workflowsclaude/fix-website-access-6FKJNclaude/gatetest-integration-hardeningclaude/github-audit-improvements-bDFr9claude/gluecron-launch-status-FoMRlclaude/hopeful-lamport-olfCTclaude/issue-to-pr-and-protectionsclaude/jolly-heisenberg-2sg1Qclaude/launch-preparation-QmTb6claude/new-session-xk1l7claude/plan-platform-architecture-kkN4yclaude/platform-analysis-roadmap-1nUGLclaude/platform-launch-assessment-8dWV8claude/polish-platform-release-AeDrUclaude/resume-previous-work-KzyLwclaude/review-crontech-handoff-qYEVqclaude/review-project-completeness-lHhS2claude/review-readme-docs-ulqPKclaude/serene-edison-rj87weclaude/setup-multi-repo-dev-BCwNQclaude/ship-fixes-and-tests-Jvz1cclaude/site-audit-competitive-pctlwgclaude/site-migration-vercel-XstpKclaude/standalone-product-repos-XHFTDcopilot/feat-smart-empty-states-keyboard-first-enhancementcopilot/feat-smart-morning-digest-review-context-restorecopilot/fix-and-process-workflowscopilot/update-ai-powered-code-reviewfeat/debt-mapfeat/push-policy-codeowners-hardeningfeat/smart-digest-contextfeat/stage-impactfeat/t1-secret-migrationfeat/u-polishfeat/w-self-hostfeat/w2-claude-configfix/agent-journey-orphan-sweepgatetest/auto-fix-1776586424172gatetest/auto-fix-1776586534814gatetest/auto-fix-1776590685143gatetest/auto-fix-1776590808199mainops/redeploy-retriggerstyle/dxt-cta-themeworktree-agent-a3377aad30d55da26worktree-agent-a7ef607b7ee1d6c74
cache-action.ts6.9 KB · 227 lines
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
/**
 * `gluecron/cache@v1` — RESTORE-only cache action (v1 scope).
 *
 * Looks up `workflow_run_cache` by (repoId, key, scope='repo') and unpacks
 * the stored tar archive into `ctx.workspace/<path>`. If no exact key hit,
 * tries each `restoreKeys` entry as a prefix match ordered by most-recently
 * used. Sets `cache-hit` output to 'true' or 'false'.
 *
 * TODO (v2): cache SAVE on job success. The deferred design is for the
 * runner to honor a `save-cache: true` flag emitted by this action and
 * call a `saveCache(ctx, key, path)` helper at end-of-job. Implementing
 * save inline here is error-prone (we'd need a post-hook) and the spec
 * explicitly endorsed shipping restore-only for v1. Size cap logic is
 * stubbed below so it's trivial to wire up later.
 *
 * Failure tolerance: any error — DB miss, tar failure, unknown scope —
 * results in `cache-hit: false` with exitCode 0. Caching is an optimization;
 * losing it must never break a pipeline.
 */

import { and, eq, isNull, sql } from "drizzle-orm";
import { mkdir } from "fs/promises";
import { join } from "path";
import { tmpdir } from "os";
import type { ActionHandler, ActionContext } from "../action-registry";
import { db } from "../../db";
import { workflowRunCache } from "../../db/schema";

// 100MB cap — reserved for the eventual save path. Kept here so both
// halves of the action share the constant when v2 lands.
export const MAX_CACHE_BYTES = 100 * 1024 * 1024;

function parseInputs(ctx: ActionContext): {
  key: string;
  path: string;
  restoreKeys: string[];
} | null {
  const w = ctx.with || {};
  const key = typeof w.key === "string" ? w.key : "";
  const path = typeof w.path === "string" ? w.path : "";
  const restoreKeysRaw = w.restoreKeys ?? w["restore-keys"];
  const restoreKeys = Array.isArray(restoreKeysRaw)
    ? restoreKeysRaw.filter((k): k is string => typeof k === "string")
    : [];
  if (!key || !path) return null;
  return { key, path, restoreKeys };
}

/**
 * Find a cache row for this repo matching either the exact key or any
 * of the prefix restoreKeys. Returns the first hit (prefix matches are
 * ordered by `last_accessed_at DESC`).
 */
async function lookupCache(
  repoId: string,
  key: string,
  restoreKeys: string[]
): Promise<
  | { hit: true; id: string; content: Buffer; matchedKey: string; exact: boolean }
  | { hit: false }
> {
  // Exact match first.
  const exact = await db
    .select()
    .from(workflowRunCache)
    .where(
      and(
        eq(workflowRunCache.repositoryId, repoId),
        eq(workflowRunCache.cacheKey, key),
        eq(workflowRunCache.scope, "repo"),
        isNull(workflowRunCache.scopeRef)
      )
    )
    .limit(1);
  const exactRow = exact[0];
  if (exactRow) {
    return {
      hit: true,
      id: exactRow.id,
      content: normalizeBytea(exactRow.content),
      matchedKey: exactRow.cacheKey,
      exact: true,
    };
  }

  // Prefix fallbacks (LRU order).
  for (const prefix of restoreKeys) {
    if (!prefix) continue;
    const rows = await db
      .select()
      .from(workflowRunCache)
      .where(
        and(
          eq(workflowRunCache.repositoryId, repoId),
          eq(workflowRunCache.scope, "repo"),
          isNull(workflowRunCache.scopeRef),
          sql`${workflowRunCache.cacheKey} LIKE ${prefix + "%"}`
        )
      )
      .orderBy(sql`${workflowRunCache.lastAccessedAt} DESC`)
      .limit(1);
    const row = rows[0];
    if (row) {
      return {
        hit: true,
        id: row.id,
        content: normalizeBytea(row.content),
        matchedKey: row.cacheKey,
        exact: false,
      };
    }
  }

  return { hit: false };
}

/**
 * `content` arrives as a Buffer, Uint8Array, or (if the driver ran through
 * text serialization) a base64/hex string. Normalize to Buffer.
 */
function normalizeBytea(raw: unknown): Buffer {
  if (Buffer.isBuffer(raw)) return raw;
  if (raw instanceof Uint8Array) return Buffer.from(raw);
  if (typeof raw === "string") {
    // Postgres `bytea` text encoding can be `\x…` hex. Handle defensively.
    if (raw.startsWith("\\x")) return Buffer.from(raw.slice(2), "hex");
    // Otherwise assume base64 (matches workflow-artifacts convention).
    try {
      return Buffer.from(raw, "base64");
    } catch {
      return Buffer.alloc(0);
    }
  }
  return Buffer.alloc(0);
}

async function unpackTar(content: Buffer, destDir: string): Promise<void> {
  await mkdir(destDir, { recursive: true });
  // Write the archive to a tmp file then untar. Piping through stdin works
  // but a tmp file avoids subtle Bun subprocess stdin EAGAIN edge cases.
  const tmpPath = join(
    tmpdir(),
    `gluecron-cache-${Date.now()}-${crypto.randomUUID().replace(/-/g, '').slice(0, 8)}.tar`
  );
  await Bun.write(tmpPath, content);
  try {
    const proc = Bun.spawn(["tar", "-xf", tmpPath, "-C", destDir], {
      stdout: "pipe",
      stderr: "pipe",
    });
    await proc.exited;
  } finally {
    try {
      await Bun.file(tmpPath).exists();
      // Best-effort cleanup; ignore errors.
      await import("fs/promises").then((fs) => fs.unlink(tmpPath)).catch(() => {});
    } catch {
      /* noop */
    }
  }
}

async function touchLastAccessed(id: string): Promise<void> {
  try {
    await db
      .update(workflowRunCache)
      .set({ lastAccessedAt: new Date() })
      .where(eq(workflowRunCache.id, id));
  } catch {
    // Non-fatal — LRU accuracy is best-effort.
  }
}

export const cacheAction: ActionHandler = {
  name: "gluecron/cache",
  version: "v1",
  async run(ctx): Promise<import("../action-registry").ActionResult> {
    // Every failure path below returns exitCode 0 with cache-hit=false so
    // the pipeline keeps flowing. This is load-bearing behaviour.
    try {
      const inputs = parseInputs(ctx);
      if (!inputs) {
        return {
          exitCode: 0,
          outputs: { "cache-hit": "false" },
          stderr: "cache: missing required inputs `key` and `path` — skipping",
        };
      }

      const result = await lookupCache(
        ctx.repoId,
        inputs.key,
        inputs.restoreKeys
      );

      if (!result.hit) {
        return {
          exitCode: 0,
          outputs: { "cache-hit": "false" },
          stdout: `cache miss for key=${inputs.key}`,
        };
      }

      const dest = join(ctx.workspace, inputs.path);
      await unpackTar(result.content, dest);
      await touchLastAccessed(result.id);

      return {
        exitCode: 0,
        outputs: {
          "cache-hit": result.exact ? "true" : "false",
          "matched-key": result.matchedKey,
        },
        stdout: `cache ${result.exact ? "hit" : "partial hit"} for key=${inputs.key} (matched=${result.matchedKey})`,
      };
    } catch (err) {
      // Fail-open: swallow and report as miss.
      return {
        exitCode: 0,
        outputs: { "cache-hit": "false" },
        stderr:
          "cache error (non-fatal): " +
          (err instanceof Error ? err.message : String(err)),
      };
    }
  },
};