fix(memory): circuit breaker for Upstash Vector search failures

[konard]

Summary

Fixes #477

When Upstash Vector backend is temporarily unavailable, the agent was logging a WARN on every semantic search attempt (which runs on every incoming message), flooding the logs with:

WARN: [Memory] Semantic Memory: Fallback Mode (Upstash Vector search failed)
    err: { "type": "UpstashError", "message": "The vector store backend is currently unavailable..." }

Root Cause

HybridSearch.semanticVectorSearchKnowledge() catches every error from UpstashSemanticVectorStore.searchKnowledge() and logs a WARN. Without any cooldown, this fires on every message when Upstash is down.

Fix

Added a 60-second circuit breaker to UpstashSemanticVectorStore.searchKnowledge():

• On the first failure, the circuit opens and the error is re-thrown (so HybridSearch logs the warning once and falls back to local keyword search).
• While the circuit is open, subsequent calls return [] immediately — no network call, no repeated log entry.
• After 60 seconds the circuit closes and the next call retries Upstash normally.
• On success, the circuit resets.
• configure() also resets the circuit breaker so reconfiguring the store starts fresh.

How to Reproduce / Verify

1. Configure UPSTASH_VECTOR_REST_URL and UPSTASH_VECTOR_REST_TOKEN with valid credentials.
2. Make the Upstash endpoint unreachable (e.g. wrong URL or network block).
3. Send multiple messages to the agent.
4. Before fix: A WARN is logged for every message.
5. After fix: The WARN fires once, then is suppressed for 60 seconds while local memory continues to work.

Tests

Added two new test cases to src/memory/__tests__/vector-store-timeout.test.ts:

• circuit breaker skips Upstash calls after a failure without re-throwing — verifies the second call within the cooldown returns [] silently without contacting Upstash.
• circuit breaker resets after the cooldown period — verifies that after 60 seconds the circuit closes and Upstash is contacted again.

All 360 memory-module tests pass.

[konard]

🤖 AI Work Session Started

Starting automated work session at 2026-05-05T17:07:44.531Z

The PR has been converted to draft mode while work is in progress.

This comment marks the beginning of an AI work session. Please wait for the session to finish, and provide your feedback.

[konard]

Working session summary

Done. Here's a summary of what was done:

Root cause: When Upstash Vector backend is unavailable, HybridSearch.semanticVectorSearchKnowledge() catches errors from the store's searchKnowledge() and logs a WARN on every semantic search — which runs on every incoming message. This floods the logs with the same error repeatedly.

Fix: Added a 60-second circuit breaker to UpstashSemanticVectorStore.searchKnowledge() (src/memory/vector-store.ts):

• First failure → circuit opens, error re-thrown (warning logged once by HybridSearch, falls back to local keyword search)
• While circuit is open → returns [] silently, no network call, no log spam
• After 60s → circuit resets, next call retries Upstash
• Success → circuit resets immediately
• configure() also resets the circuit

Two new tests verify the circuit opens on failure and resets after the cooldown. All 360 memory-module tests pass.

PR: #478

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $1.125849

📊 Context and tokens usage:

• 75.0K / 1M (7%) input tokens, 12.0K / 64K (19%) output tokens

Total: (44 new + 75.3K cache writes + 2.2M cache reads) input tokens, 12.0K output tokens, $1.125849 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: sonnet
• Model: Claude Sonnet 4.6 (claude-sonnet-4-6)

📎 Log file uploaded as Gist (1341KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag