Resolved incident

Degraded performance for requests using Slam-1

AssemblyAI · Minor · Started

Duration
6h 45m
Severity
Minor
Status
Resolved

Summary

Minor · Resolved

We have continued to see good performance and success rates for Slam-1 and are now marking this issue as resolved.

View official incident ↗

Event timeline

How this incident unfolded, from the official status page.

  1. Resolved

    AssemblyAI

    We have continued to see good performance and success rates for Slam-1 and are now marking this issue as resolved.

  2. Monitoring

    AssemblyAI

    Success rates and processing times have returned to the normal range, but we are monitoring traffic to ensure continued good performance.

  3. Identified

    AssemblyAI

    We are once again seeing occasional issues where some requests using the Slam-1 model are taking longer than normal to process or failing with a server error. Retries are typically successful and requests using the Universal model are unaffected.

  4. Monitoring

    AssemblyAI

    We are continuing to see success rates in the normal range and no longer than usual processing times but we will continue to monitor traffic to ensure continued good performance.

  5. Identified

    AssemblyAI

    We are starting to see success rates return to the normal range, but we are still working to ensure that the issue has been fully resolved.

  6. Identified

    AssemblyAI

    We are currently seeing some Slam-1 requests fail with a server error as we are restarting services while working to address this issue. We suggest holding off on retries for the time being or retrying those requests using Universal until normal processing has been restored. We will release another update here as soon as we are seeing an increase in success rates.

  7. Identified

    AssemblyAI

    Our team has identified the issue causing the delay in processing requests using Slam-1 and is actively working to resolve the issue.

  8. Investigating

    AssemblyAI

    We are currently investigating an issue that is leading to longer-than-usual processing times for pre-recorded audio transcriptions using the Slam-1 model. Requests using the Universal model, which is enabled by default, are not affected by this issue.