Visitar URL original
Repeated out-of-memory crash during a long interactive session in 1.0.109 · Issue #5435 · github/copilot-cli · GitHub
Skip to content

Repeated out-of-memory crash during a long interactive session in 1.0.109 #5435

Open

Description

@paulrigor

Description

Copilot CLI crashed again with an out-of-memory condition during a long interactive session. The user reports this is recurrent. After the crash the session had to be resumed, and the previous background agent was no longer accessible from the resumed session.

Environment

  • Copilot CLI: 1.0.109
  • OS: Linux (Rocky Linux 9)
  • Workload: long interactive session with prior context compaction and background agent work

Expected

The CLI remains responsive, or recovers without losing access to in-progress work.

Actual

The CLI crashed with an out-of-memory condition. Exact allocation, stack trace, trigger, and reproducible steps have not been independently established; this report does not assert a root cause. No diagnostic logs are included.

Possibly related: #4699 and #2132.

Activity

  1. paulrigor commented on Nov 12, 2026

    @paulrigor
    Author

    Another recurrence reported on 2026-11-12 at approximately 14:17 PST (22:17 UTC), during the same long-running interactive session.

    User-supplied V8 GC diagnostics, verbatim:

    [2701543:0x16ba8000] 24360855 ms: Scavenge 3911.4 (4076.3) -> 3908.6 (4076.3) MB, pooled: 0 MB, 2.82 / 0.02 ms  (average mu = 0.972, current mu = 0.974) task;
    [2701543:0x16ba8000] 24361742 ms: Mark-Compact 4044.8 (4211.6) -> 3957.2 (4180.2) MB, pooled: 12 MB, 495.17 / 0.16 ms  (average mu = 0.916, current mu = 0.822) allocation failure; scavenge might not succeed
    

    The diagnostics show roughly 3.96 GB remaining after a major collection at approximately 6 hours 46 minutes of process uptime. They are consistent with substantial retained heap and allocation pressure, but do not establish a leak, the exact heap limit, or a root cause. No actual core file, fatal-error line, or native stack trace was supplied.

    Recovery observation: the resumed assistant could recover persistent task records and files, but three known in-flight background-agent IDs were no longer found, and no prior polling shell remained active. Independently running application containers remained healthy; their health does not diagnose the CLI crash.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions