Troubleshooting habit

Confirm before relying. Quotas, prices, and names change.

Leg, id, counter, capture

Which hop failed: carrier, flow, Lambda, queue, or softphone? Which ContactId joins the logs? Which counter moved: queue depth, concurrent calls, Lambda errors, or UDP failures? Which capture proves it: flow log, CTR, endpoint test, or a recording?

Change one thing. Retest that ContactId's scenario. Then stop.

• A flow log error with no Lambda log is a permission or a payload rejection.
• A CCP error that names TurnNlb is the network, even if the flow is perfect.
• A fast busy on a campaign is the zero quota until proven otherwise.
• A missing recording is the block, the bucket, or the key.

Source: re:Post, flow errors

Faults worth memorizing

Softphone text that names TurnNlb is UDP 3478 or DNS. Lambda AccessDenied is the resource policy or the association. A 10-second function with an 8-second block is a timeout even when Lambda 'succeeded'. Campaign fast busy is the zero quota. The first phone-number claim can fail at quota zero-looking errors that need Support.

Voice ID faults after 20 May 2026 are expected. Remove the blocks. Email threads split at 90 days. Idle chats fill the 500.

• SAML users with no routing profile signed in and cannot take work.
• Global Resiliency numbers missing from the replica were never in a distribution group.
• A recording missing from S3 is policy, key, or the block.
• CTR delay under five minutes is buffering. Past that, check Firehose errors.

Source: CCP issues