You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Windows 26.928.2636.0: Repeated renderer V8 OOM crashes reload the UI while background tasks continue #49666
Environment
Windows 11 x64, 32 GB RAM.
Application 26.928.21956, build 12404, prod.
Microsoft Store package OpenAI.Codex 26.928.2636.0.
Symptoms
Every few minutes, the UI suddenly disappears, shows a loading icon, and returns to the home page; the main process and some background tasks continue running, while the Dot voice call disconnects. The user reports that this can happen even without clicking the chat again.
A 14-second recording shows that after switching from a chat with work in progress to another chat, the entire UI starts reloading around seconds 4–5 and returns to the home page around second 9. The recording has no clock that can be cross-checked, so it has not been forcibly matched to a specific crash timestamp.
Confirmed native crash data
A renderer minidump was captured at 2026-09-30 23:24:08 UTC+08:00:
Exception code 0xe0000008.
ptype=renderer.
v8-oom-location=CALL_AND_RETRY_LAST.
v8-oom-last-few-messages includes allocation failure, GC in old space requested, and last resort.
After the last two garbage collections, the JavaScript heap was approximately 2562.3 MB and 2561.5 MB; allocation still failed.
This confirms that the immediate failure in that instance was a V8 memory allocation failure, but the specific JavaScript caller has not been identified.
Other observations
Another renderer crash sidecar was created at 23:42:06.509; a sample taken at 23:42:06.171 showed that the UI process had 4578.6 MiB of private memory and a 4085.6 MiB working set, with overall system RAM usage at 63.5%. Exhaustion of the system's physical RAM cannot explain this observation.
Another renderer crash sidecar was created at 23:44:52.911. These two sidecars do not themselves contain exception codes, so it has not been assumed that all crashes have the same native stack.
For the same large local chat, thread/items/list responses have at times reached 300–500+ items per minute. The chat history file is approximately 38.61 GiB; this is its size on disk, not the UI's memory usage or the size of each response.
History responses were still observed after switching away from the chat; duplicate cursors have not been proven, and a controlled minimal reproduction has not yet been established.
Codex Doctor checks for installation, settings loading, the desktop app-server handshake, HTTP/WebSocket reachability, and the database integrity checks performed all passed. The general warnings in the diagnostics have not been shown to be related to this OOM.
Related public cases #49255 #46690 #48938 #47793
These are leads for comparison; there is no claim that they all share the same root cause.
Requested assistance
Confirm whether this is related to known Windows renderer/history pagination issues.
Provide a supported temporary workaround that preserves running tasks and chat data.
Identify the specific version containing a fix, or specify any additional diagnostic data needed and a secure submission channel.
Privacy
This summary contains no account details, email addresses, chat identifiers, project names, file paths, conversation content, or credentials. The native dump and recording remain local and have not been made public; the native dump may contain process memory contents.
I can reproduce a severe Windows desktop memory-growth problem that does not require the originally suspected project/task.
Reproduction / timeline
Windows 11 desktop app. I initially suspected one specific task because memory usage exploded while it was running, but a controlled follow-up showed the same behavior with an unrelated task.
Approximate observations from Task Manager screenshots taken on 2026-10-01 (local time UTC+05:30):
~13:13 — Codex process group: ~1.5 GB RAM, system memory ~55%.
~13:18 — after settling / before the unrelated workload became heavy: ~1.24 GB.
~13:19 — one unrelated Codex task running: ~2.70 GB, system memory ~70%.
~13:20 — same task: ~2.51 GB.
~13:21 — same task: ~3.08 GB.
~13:40 — only one Codex task running, Codex process group: ~9.95 GB RAM, system memory 97%.
At ~13:40 Codex was also using ~29% CPU. The central conversation pane was stuck on a loading spinner while the task card still indicated activity.
Why this seems important
The original suspected task is not required to reach the pathological memory state. A different task caused the Codex desktop process group to grow from roughly 1.2–1.5 GB to nearly 10 GB over ~20 minutes.
This makes an app/renderer/session memory problem much more likely than a project-specific workload alone. It may also be related to the UI/chat-loading failure: while memory was high, the conversation view repeatedly remained on a spinner.
Additional observations
Chrome and other applications were relatively small compared with Codex at the 97% system-memory point.
Task Manager showed the Codex desktop group itself as the dominant consumer.
The problem persisted with just one visible Codex task active.
I have multiple timestamped Task Manager screenshots demonstrating the progression.
Privacy / evidence
I am not uploading the raw screenshots here because they contain account/project-identifying information. I can provide redacted screenshots or additional diagnostics if maintainers specify what would be most useful.
Request
Could maintainers confirm:
whether this matches the known Windows renderer/V8 OOM issue;
whether there is a supported workaround or rollback target;
what diagnostics would best distinguish renderer/history-loading leakage from task-execution memory growth; and
which build contains a fix when one is available?
This report is intended as additional reproduction evidence for the same class of Windows desktop OOM/memory-growth issue.
Activity
[-]Windows 26.928.2636.0 反覆 renderer V8 OOM,介面重新載入而背景任務繼續[/-][+]Windows 26.928.2636.0: Repeated renderer V8 OOM crashes reload the UI while background tasks continue[/+]github-actions commentedon Sep 30, 2026
English translation:
Environment
Windows 11 x64, 32 GB RAM.
Application 26.928.21956, build 12404, prod.
Microsoft Store package OpenAI.Codex 26.928.2636.0.
Symptoms
Every few minutes, the UI suddenly disappears, shows a loading icon, and returns to the home page; the main process and some background tasks continue running, while the Dot voice call disconnects. The user reports that this can happen even without clicking the chat again.
A 14-second recording shows that after switching from a chat with work in progress to another chat, the entire UI starts reloading around seconds 4–5 and returns to the home page around second 9. The recording has no clock that can be cross-checked, so it has not been forcibly matched to a specific crash timestamp.
Confirmed native crash data
A renderer minidump was captured at 2026-09-30 23:24:08 UTC+08:00:
Exception code 0xe0000008.
ptype=renderer.
v8-oom-location=CALL_AND_RETRY_LAST.
v8-oom-last-few-messages includes allocation failure, GC in old space requested, and last resort.
After the last two garbage collections, the JavaScript heap was approximately 2562.3 MB and 2561.5 MB; allocation still failed.
This confirms that the immediate failure in that instance was a V8 memory allocation failure, but the specific JavaScript caller has not been identified.
Other observations
Another renderer crash sidecar was created at 23:42:06.509; a sample taken at 23:42:06.171 showed that the UI process had 4578.6 MiB of private memory and a 4085.6 MiB working set, with overall system RAM usage at 63.5%. Exhaustion of the system's physical RAM cannot explain this observation.
Another renderer crash sidecar was created at 23:44:52.911. These two sidecars do not themselves contain exception codes, so it has not been assumed that all crashes have the same native stack.
For the same large local chat, thread/items/list responses have at times reached 300–500+ items per minute. The chat history file is approximately 38.61 GiB; this is its size on disk, not the UI's memory usage or the size of each response.
History responses were still observed after switching away from the chat; duplicate cursors have not been proven, and a controlled minimal reproduction has not yet been established.
Codex Doctor checks for installation, settings loading, the desktop app-server handshake, HTTP/WebSocket reachability, and the database integrity checks performed all passed. The general warnings in the diagnostics have not been shown to be related to this OOM.
Related public cases
#49255
#46690
#48938
#47793
These are leads for comparison; there is no claim that they all share the same root cause.
Requested assistance
Confirm whether this is related to known Windows renderer/history pagination issues.
Provide a supported temporary workaround that preserves running tasks and chat data.
Identify the specific version containing a fix, or specify any additional diagnostic data needed and a secure submission channel.
Privacy
This summary contains no account details, email addresses, chat identifiers, project names, file paths, conversation content, or credentials. The native dump and recording remain local and have not been made public; the native dump may contain process memory contents.
github-actions commentedon Sep 30, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
ankiygoescrazy commentedon Oct 1, 2026
I can reproduce a severe Windows desktop memory-growth problem that does not require the originally suspected project/task.
Reproduction / timeline
Windows 11 desktop app. I initially suspected one specific task because memory usage exploded while it was running, but a controlled follow-up showed the same behavior with an unrelated task.
Approximate observations from Task Manager screenshots taken on 2026-10-01 (local time UTC+05:30):
At ~13:40 Codex was also using ~29% CPU. The central conversation pane was stuck on a loading spinner while the task card still indicated activity.
Why this seems important
The original suspected task is not required to reach the pathological memory state. A different task caused the Codex desktop process group to grow from roughly 1.2–1.5 GB to nearly 10 GB over ~20 minutes.
This makes an app/renderer/session memory problem much more likely than a project-specific workload alone. It may also be related to the UI/chat-loading failure: while memory was high, the conversation view repeatedly remained on a spinner.
Additional observations
Privacy / evidence
I am not uploading the raw screenshots here because they contain account/project-identifying information. I can provide redacted screenshots or additional diagnostics if maintainers specify what would be most useful.
Request
Could maintainers confirm:
This report is intended as additional reproduction evidence for the same class of Windows desktop OOM/memory-growth issue.