[superlog] Downgrade agent stream cleanup Redis timeouts from ERROR to WARN

[superlog-app]

Summary

Transient Redis timeouts during agent chat stream cleanup (specifically clearActiveStream, markStreamDone, and appendStreamChunk) were being logged at ERROR level, generating false-positive incidents. The primary AI response is already sent to the client before these background operations run — they only affect the Redis stream buffer used for client reconnects.

When a Redis command times out in the onFinish callback or in the background storage-writer task, captureError has no active request logger context and falls through to log.error() unconditionally. This is the correct path for errors that affect the user-visible response, but these three catch blocks are pure cleanup/buffering side-effects.

Replaces the three captureError calls in post-response cleanup paths with log.warn() (already imported). The stream replay/reconnect path remains the only potentially degraded feature (stale active-stream key for up to 1 hour on timeout), but the chat response itself is unaffected. An alternative approach would be to add a severity option to captureError so callers can choose warn vs error per call-site — that may be worth doing as a follow-up if more similar patterns appear in other routes.

Incident on Superlog

---

Was this PR helpful? Leave feedback — goes straight to the Superlog team.

---

Summary by cubic

Downgraded logging of transient Redis timeouts during agent chat stream cleanup from ERROR to WARN to prevent false-positive incidents. Chat responses are unaffected; only the reconnect buffer may be briefly stale.

• Bug Fixes
  • Replaced captureError(...) with log.warn(...) in post-response cleanup and background storage-writer paths (clearActiveStream, markStreamDone).
  • Preserved structured fields (service, error_message, agent_stream_cleanup_error/agent_stream_persist_error, agent_chat_id) for observability.
  • Eliminates noisy ERRORs triggered without a request logger context in onFinish/background tasks.

^{Written for commit 4b9ef10. Summary will update on new commits.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
databuddy-status	Ready	Preview, Comment	Jun 27, 2026 5:52am

2 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
dashboard	Skipped		Jun 27, 2026 5:52am
documentation	Skipped		Jun 27, 2026 5:52am

[unkey-deploy]

The latest updates on your projects. Learn more about Unkey Deploy

Name	Status	Preview	Inspect	Updated (UTC)
api (preview)	Ready	Visit Preview	Inspect	Jun 27, 2026 5:52am

[greptile-apps]

Greptile Summary

Downgrades three captureError calls in post-response Redis cleanup paths (clearActiveStream, markStreamDone, and the storage-writer IIFE catch) to log.warn, eliminating false-positive error alerts for transient timeouts that don't affect the already-delivered chat response.

• The old captureError calls were falling through to log.error() at module scope because no active request logger exists in the onFinish callback or in the background storage-writer task — the new log.warn calls carry the same fields (service, error_message, context keys) without changing what is captured.
• Database persistence errors (agent_persist_error) are correctly left at ERROR level via captureError, so genuinely user-visible failures still alert.

Confidence Score: 5/5

Safe to merge — the change only touches post-response background cleanup paths and leaves all user-visible error handling intact.

The replaced captureError calls were already falling through to the no-request-logger branch, which logs err.message via log.error. The new log.warn calls carry the same fields, so no diagnostic information is dropped. DB persistence errors and all user-facing error paths are correctly left untouched. The one nuance is that the storage-writer IIFE's outer .catch now suppresses stream-reader errors at WARN in addition to Redis timeouts, but this is a low-probability path.

No files require special attention — the single changed file has well-understood boundaries between user-visible and background operations.

Important Files Changed

Filename	Overview
apps/api/src/routes/agent.ts	Three captureError calls in background Redis cleanup paths replaced with log.warn; DB persistence error handling and all user-visible error paths are unchanged.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant Client
    participant AgentRoute as Agent Route
    participant Redis
    participant DB

    AgentRoute->>Client: stream forClient (response sent)
    
    par Background storage writer
        AgentRoute->>Redis: appendStreamChunk(streamKey, chunk)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
        AgentRoute->>Redis: markStreamDone(streamKey)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
    and onFinish callback
        AgentRoute->>Redis: clearActiveStream(streamScope, chatId, streamId)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
        AgentRoute->>DB: upsert agentChats (messages, title)
        note over AgentRoute,DB: DB error → captureError/log.error (unchanged)
    end

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant Client
    participant AgentRoute as Agent Route
    participant Redis
    participant DB

    AgentRoute->>Client: stream forClient (response sent)
    
    par Background storage writer
        AgentRoute->>Redis: appendStreamChunk(streamKey, chunk)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
        AgentRoute->>Redis: markStreamDone(streamKey)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
    and onFinish callback
        AgentRoute->>Redis: clearActiveStream(streamScope, chatId, streamId)
        note over AgentRoute,Redis: Redis timeout → log.warn (was captureError/log.error)
        AgentRoute->>DB: upsert agentChats (messages, title)
        note over AgentRoute,DB: DB error → captureError/log.error (unchanged)
    end

Loading

Comments Outside Diff (1)

1. apps/api/src/routes/agent.ts, line 1134-1147 (link)

   IIFE .catch silences non-Redis stream errors at WARN

   The .catch((storageError) => …) block catches everything that escapes the storage-writer IIFE, not just Redis operations. This includes reader.read() failures on the teed ReadableStream, which are unusual and typically signal a programming error rather than a transient Redis timeout. If the ReadableStream itself errors (e.g. due to a bug in createAgentUsageInjector), the symptom would now only appear as a WARN with no stack, making it hard to distinguish from the expected Redis timeout noise the PR is trying to suppress.

_{Reviews (1): Last reviewed commit: "[superlog] Downgrade agent stream cleanu..." | Re-trigger Greptile}