Request Monitoring Troubleshooting

When Dashboard or Monitoring is empty, do not start by refreshing the page or changing filters. First locate the broken link: CPA publishing, Manager Server collection, SQLite insertion, or frontend display.

Quick Checks

1. Confirm CPA has produced real requests.
2. Confirm CPA usage publishing is enabled.
3. Open Manager Server /status and check collector, lastConsumedAt, lastInsertedAt, and lastError.
4. Confirm no other Manager Server is consuming the same CPA queue.
5. Confirm pollIntervalMs is less than or equal to CPA usage queue retention.

/status Fields

Field	Meaning	How to read it
collector / mode	Current collector state and mode.	Should match USAGE_COLLECTOR_MODE or panel configuration.
queue	CPA usage queue name.	Usually usage by default.
lastConsumedAt	Last time an event was read from the CPA queue.	Empty for a long time means no event has been consumed.
lastInsertedAt	Last time an event was written to SQLite.	If events are consumed but not inserted, check lastError and skipped counts.
totalInserted	Total inserted events.	Growth means the collection path is working.
totalSkipped	Total skipped events.	Can be duplicate, invalid, or already processed events.
lastError	Latest error.	Use it first to locate network, auth, or data format issues.

Collection Modes

USAGE_COLLECTOR_MODE=auto tries:

1. RESP Pub/Sub.
2. HTTP queue.
3. RESP pop.

Behind a reverse proxy:

• RESP Pub/Sub and RESP pop must connect directly to the CPA API port, usually 8317.
• HTTP reverse proxies cannot proxy RESP.
• HTTP queue can go through an HTTP proxy.
• If only an HTTP proxy path is available, confirm CPA is at least v6.10.8+.

Retention And Polling

CPA usage queue retention defaults to 60 seconds and is capped at 3600 seconds. pollIntervalMs cannot exceed retention.

Typical problems:

• Manager Server was stopped longer than retention, so old events expired and cannot be recovered.
• pollIntervalMs is too large, so events expire before the next poll.
• Multiple Manager Servers consume the same queue and one instance removes events first.

Common Symptoms

The panel is empty and lastConsumedAt is also empty

Check:

• Whether CPA usage publishing is enabled.
• Whether the CPA URL is reachable from Manager Server.
• Whether the CPA Management Key is correct.
• Whether collector mode matches the network path.

lastConsumedAt has a value, but lastInsertedAt does not change

Check:

• Whether lastError contains SQLite, JSON, or field compatibility errors.
• Whether the SQLite data directory is writable.
• Whether restored usage.sqlite and data.key do not match.

Events are missing intermittently

Check:

• Whether retention is too short.
• Whether Manager Server is restarted frequently.
• Whether multiple Manager Servers consume the same queue.
• Whether CPA and Manager Server clocks have a large skew.

Fix Order

1. Let Manager Server connect directly to CPA :8317 first to rule out proxy issues.
2. Temporarily set USAGE_COLLECTOR_MODE to auto.
3. Increase CPA usage queue retention and keep pollIntervalMs lower.
4. Ensure only one Manager Server consumes the same CPA queue.
5. If monitoring is still empty, keep /status, Manager Server logs, and CPA usage configuration for further diagnosis.