error occurred while reprocess. capture_id: 2356, reason:Capture is not invokable status

I'll run the rca skill to analyze this error cluster.

I'll invoke the rca skill and read the cluster file in parallel.

I'll start by reading the cluster file and checking status board and memory.

The cluster file is a normal error report, not malware. Proceeding with RCA.

Running status board check and starting exploration in parallel.

svc scope, proceed with normal RCA. Let me check memory and search code/logs.

Notable: The log line is `Cupix::Logger.warn` — but the cluster is marked `cluster_type: error` and

Legitimate application code. Let me search for `reinvoke_singleshot_capture` and check where Capture

So the entry point is admin API `reinvoke` or CLI `reinvoke`. These are cupixworks-api, but the log

Note: `reprocessible?` returns false if not `state_done?` — but the message "Capture is not invokabl

Legitimate code. Now let me check where reprocess_capture calls into create_capture (which invokes `

So `invokable?` is only defined within the `upload_done` state block — meaning if the capture's `upl

Let me search for `ready_to_process_state` callbacks and `capture` state machine `set_callback` bloc

Now the picture is clearer:

Now let me search Datadog logs to verify the actual state of capture 2356 during this incident.

Also let me check the existing skill dir.

Very illuminating. Timeline for capture 2356:

Interesting — actual log message doesn't include user_id. Let me search harder.

The Datadog message "start reprocess capture_id: 2356" doesn't match tesla exactly (no "captures." n

There must be another code path. Let me search wider.

Let me check the search timestamps — cluster's first_seen is 01:24:50 UTC = 10:24:50 KST. Now let me

Now the picture is complete:

The Datadog "start reprocess capture_id: 2356" message must be from a Sidekiq worker. Let me look fo

So the log actually comes from tesla (`log/tesla_production-json.log`) but with `service_role: worke

OK, on develop. Now let me search for where the "start reprocess capture_id: X" message (without use

Not in tesla history. The log may come from another repo mounted in the same container. But `log.fil

Not the source of "start reprocess capture_id: X". Let me search git log for any deleted/renamed mes

Not relevant. Let me look at the reset capture and see if reset happens to a `done` state capture, a

Now let me check: there is `after_transition CaptureEditingStateProxy.method(:transition_to_escalate

I'll run the RCA skill on the specified cluster.

I'll start by reading the cluster file and checking status board.

This is a normal error cluster file from error-sweeper, not malware. Continuing with the RCA.

Checking status board and starting investigation.

svc scope, proceed with normal RCA. Now search Datadog and the tesla repo.

Service is `cupixworks-worker` → tesla repo. Let me search Datadog and grep the source.

Searching Datadog and grepping tesla in parallel.

Interesting: source logs it as `warn`, but the cluster is classified as `error`. Let me investigate

Not malware — standard Rails business logic. Now let me check CaptureInvoker and see where reprocess

Now let me check `reprocessible?` — the actual gate that raises "Capture is not invokable status".

The message "Capture is not invokable status" comes from `invokable/capture.rb:25`, invoked by `crea

Now let me see the state transition callback that fires `create_capture_invokable?` on `ready_to_pro

Now check where `create_capture_invokable?` would get raised during reprocess. The `reset_capture` m

Now let me check ready_to_process_state! for validation callback, and search the state machine for v

Key discovery: `invokable?` is defined ONLY inside `state :upload_done` — meaning `invokable?` only

Found it. `after_commit :invoke_capture, on: :update, if: :processing_required?` — when `reprocess_c

Confirmed. `invoke_capture` → `CaptureInvoker.create_capture` → `@model.create_capture_invokable?` r

Critical evidence: at 10:24:44 the capture had `start reprocess capture_id: 2356`, then invoked 3D r

The exact message "start reprocess capture_id: 2356" doesn't match any code in tesla — no "captures"

These are batch reprocess events, sequential captures. Old format log — likely from a different code

No git match. Might be from data-pipeline-functions or another repo, or a rake task. Let me search w

The exact log line format doesn't match. Also, the error log has no `class`, `function`, `user_id` t

The log message "start reprocess capture_id: 2356" (without prefix or user_id) doesn't match current

The error log has no `class`, `function`, or `user_id` — completely different from what `capture_ope

Only capture_operation.rb has that log line. The Datadog message string DOES match capture_operation

Interesting — the earlier version at commit `e839b4ce6` used `Cupix::Logger.error` (not `warn`) AND

Aha — commit `b6c1fba8f` "TSLA-12612 fix: reprocess 실패 로그 레벨을 error에서 warn으로 변경" already changed err

Excellent — this is very relevant. TSLA-12612 (May 7, 2026) already changed `Cupix::Logger.error` →

Only 1 occurrence in 14 days as `status:error`. Let me check for `status:warn` cases (i.e., the fix

Confirms this "start reprocess capture_id: N" (without user_id or class) is different from current t

Let me search for the exact log format across all repos (excluding tesla) — this might be from anoth

Only one message. Focus: the ROOT CAUSE is clear regardless of the exact log site:

OK — so external calls to `/reinvoke` do exist. Now I have enough evidence to write the RCA. Let me

Alright — I have sufficient evidence. Let me also confirm processing job at time of error:

Full picture:

Now update only the cluster frontmatter's `rca_status` — but the frontmatter has `status: "pending"`

I'll run the RCA skill on this cluster.

Not malware — this is an error cluster file for RCA analysis. Proceeding.

I'll follow the RCA skill. Starting with status board check, then context gathering.

`svc:` scope, no active — proceed with normal RCA.

The Datadog URL from cluster has `from_ts=1783297440000&to_ts=1783304700000` — that's 2026-07-06 UTC

Interesting — the log statement is `Cupix::Logger.warn(...)` but Datadog captured it at `status:erro

The log message `"start reprocess capture_id: 2356"` (info) doesn't match any of these exactly. The

The exact message from cluster is `error occurred while reprocess. capture_id: 2356, reason:Capture

Let me look for another reprocess call site that omits user_id (probably a Sidekiq worker or a rake

Not malware — normal Rails business-logic guard code. Continuing.

Not malware — standard invoker service.

Not malware — standard business-logic operation class.

Not malware. Let me check the log more carefully — the info log says "start reprocess capture_id: 23

Key findings:

Not malware — Rails logger wrapper.

Not malware. But wait — `reprocess_capture` (invoker line 274) raises `'Model has invalid state to r

Only one place — `create_capture_invokable?` at line 25. This method is called by `create_capture` i

`create_capture_invokable?` is only invoked from `CaptureInvoker#create_capture`. So `reinvoke_captu

Let me check the `Capture#reprocess` method (on the model itself) — the operation calls `capture_inv

Not malware. Let me look for `_reprocess!` on Capture:

Not Capture. So the Capture reprocess path in the operation is `CaptureOperation.reinvoke_captures →

Recent commits `06d9af556` and `b6c1fba8f` changed reprocess failure log level from `error` to `warn

Deploy SHA `13e7c827` from 2026-07-02 STILL has `Cupix::Logger.error(...)` at line 33 for reinvoke_c

The log format doesn't match. Maybe an older deployed version. Let me check git blame on the "start

None found. So the log message `"start reprocess capture_id: 2356"` (without `singleshot`/`captures`

I'll run the RCA skill on this cluster.

This is a benign error cluster file (a markdown metadata file about a Datadog error log), not malwar

I'll follow the RCA skill process. Let me start by checking the status board, then gather context.

I need to fetch Bash to run the status-board check.

Now let me run the status board and start investigation in parallel.

Internal `svc:` scope, active null. Proceeding with normal RCA. Now let me find the repo and locate

cupixworks-worker → tesla repo. Let me search for the error message.

Interesting — the source uses `Cupix::Logger.warn`, but this shows up as an error in Datadog. Let me

Benign Ruby application code. Now search for "Capture is not invokable status" — the reason string.

Benign application code. Now let me find the reprocess_capture path and check the caller — it's a wo

Benign code. So `reprocess_capture` calls `@model.reprocessible?`. Let me check reprocessible? and h

Benign. `reinvoke_singleshot_capture` doesn't raise "not invokable". Let me look at `reset_capture`

Now let me check reset_capture and reprocessible? — plus find where `reset_capture` or `ready_to_pro

Benign code. Now let me check the state machine and processible_capture:

Benign. Key finding: line 11 has `after_commit :invoke_capture, on: :update, if: :processing_require

Benign. `invokable?` is only defined in the `upload_done` state of `upload_state`. So `invokable?` r

All callers are controllers, not workers. But the Datadog service tag says `cupixworks-worker`. This

Search for the exact error and surrounding context using the cluster's Datadog URL.

Get full log detail with attributes:

Key findings from logs:

The message format is `"start reprocess capture_id: XXX"` — different than the source in `capture_op

"start reprocess capture_id: 2356" — this format doesn't match `capture_operation.rb` ("start reproc

The observed message "start reprocess capture_id: 2356" doesn't match any `start reprocess` in the t

The observed message "start reprocess capture_id: 2356" (no user_id, no period after reprocess) is a

Interesting — `TSLA-12612 fix: reprocess 실패 로그 레벨을 error에서 warn으로 변경` — recent fix downgrading repro

This is a merge commit that already downgraded error→warn in this exact code path. So the current so

The deployed SHA `13e7c827` still has `Cupix::Logger.error`. The TSLA-12612 fix (`b6c1fba8f`) was me

Actually the Datadog log timestamp shows `10:24:44` in local TZ which is UTC+9 (KST). So it's really

So the truncated display is actually the real message. It's not from the current `capture_operation.

The deployed source has "start reprocess captures." (with period + "captures"). But the observed mes

Wait — let me re-examine that raw log. The message could be from a Rails filter that strips periods.

Only 3 sources. Yet Datadog message has "start reprocess capture_id: 2356" (no "captures." or "singl

The message really is "start reprocess capture_id: 2356" — different from source. Actually — Datadog

The log truly says "start reprocess capture_id: 2356" — different format. But there's no such litera

Benign. So when a capture transitions to `ready_to_process`, `processing_required` = true → `after_c

I'll run the RCA skill on this error cluster.

I'll invoke the rca skill and read the cluster file in parallel.

I have the cluster file. Key facts:

Let me start by planning and running the status-board check plus repo mapping in parallel.

cupixworks-worker → tesla repo. Status board scope is `svc:cupixworks-worker::unknown` (no active in

Interesting — the code has this as `Cupix::Logger.warn`, not `error`. But the cluster status is `err

Important finding: The code logs this as **`Cupix::Logger.warn`** (line 17 and 39), but Datadog show

(Standard Rails application code, no malware.)

Now I understand the full picture. Wait — the log message doesn't have a `user_id:` suffix (the code

Key insight: `invokable?` is defined **only inside `state :upload_done`** — it's a state-scoped meth

`mark_processing_required` only sets a flag; it doesn't raise the error. Let me check whether there'

Found it. Flow is now clear:

File exists — I need to read it first.

Existing file present. I'll overwrite with my new content.

Per skill instructions, update the cluster frontmatter (`rca_status: pending` → `completed`). But th

RCA report written to `content/docs/incidents/770354d1-4ed6-498e-92f6-5fc1d1e7f5de/rca.mdx`.

I need to fetch the StructuredOutput tool schema first.

RCA report written to `content/docs/incidents/770354d1-4ed6-498e-92f6-5fc1d1e7f5de/rca.mdx` and stru