macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0

Summary

After upgrading to rust-v0.142.0, I can still reproduce persistent SQLite log churn on macOS in ~/.codex/logs_2.sqlite.

This looks like a partial fix:

#29432 appears to have helped: codex_api::endpoint::responses_websocket is much lower than before.
However, #29457 does not appear to fully solve the persistent SQLite churn in this app build/runtime. The release note says noisy persistent log targets were filtered, and the PR says bridged target=log events should be excluded from the SQLite sink, but TRACE target=log is still the dominant persisted target here.

I also posted an earlier data point in the closed umbrella issue #28224 here:

#28224 (comment)

Opening a new issue because #28224 is closed and this is post-rust-v0.142.0 macOS data.

Environment

Platform: macOS 26.5.1, arm64
Codex app: 26.616.71553, build 4265
Embedded CLI: codex-cli 0.142.0
PATH CLI: codex-cli 0.142.0
App-server process: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
App-server PID/start at sample time: 7763, started Tue Jun 23 10:17:10 2026
RUST_LOG: empty
~/.codex/config.toml: [analytics] enabled = false
SQLite trigger count during the sample: 0
logs_2.sqlite journal mode: wal

No local block_log_inserts trigger was present during this sample.

Files observed

At the start of the sample:

~/.codex/logs_2.sqlite      745,517,056 bytes
~/.codex/logs_2.sqlite-wal    4,437,272 bytes
~/.codex/logs_2.sqlite-shm       32,768 bytes

The active app-server process was holding the DB/WAL/SHM files open.

Read-only 60 second sample

Sample window:

2026-06-23 10:43:00 +0800 to 2026-06-23 10:44:00 +0800

SQLite metadata before/after:

trigger_count:        0 -> 0
row_count:       200589 -> 200589
max_id:       725810463 -> 725811933
sqlite_seq:   725810463 -> 725811933
max_ts:      1782182580 -> 1782182638

So the retained row count stayed flat, but sqlite_sequence / max_id advanced by 1,470 in about 60 seconds. This looks like continued insert-and-prune churn.

File mtimes also moved during the sample:

logs_2.sqlite mtime:     1782182580 -> 1782182613
logs_2.sqlite-wal mtime: 1782182580 -> 1782182638

The WAL size stayed stable at about 4.4 MB, so file size alone hides the ongoing write activity.

Recent persisted targets

Rows in the last 60 seconds:

TRACE log                                      989 rows, 526.8 KiB
TRACE codex_api::sse::responses                14 rows,  14.0 KiB
TRACE codex_app_server::outgoing_message        12 rows,   2.0 KiB
TRACE codex_mcp::connection_manager              6 rows,   4.9 KiB
DEBUG codex_core::stream_events_utils            3 rows,   3.3 KiB
INFO feedback_tags                               3 rows,   3.0 KiB
INFO codex_core::stream_events_utils             2 rows,   5.5 KiB
TRACE codex_api::endpoint::responses_websocket   1 row,   40.8 KiB
TRACE codex_core::session::turn                  1 row,    0.9 KiB

TRACE targets in the last 5 minutes:

log                                      989 rows, 526.8 KiB
codex_api::sse::responses                238 rows, 239.5 KiB
codex_mcp::connection_manager             18 rows,  13.8 KiB
codex_app_server::outgoing_message         12 rows,   2.0 KiB
codex_core::session::turn                   2 rows,   1.8 KiB
codex_api::endpoint::responses_websocket    2 rows,  80.0 KiB

Expected behavior

After rust-v0.142.0, Codex should not continuously persist high-frequency TRACE target=log rows into logs_2.sqlite by default, especially after #29457.

Actual behavior

Codex app-server continued to insert high-frequency TRACE target=log rows into the persistent SQLite log sink. The retained row count stayed flat, but max(id) / sqlite_sequence kept advancing and the DB/WAL mtimes kept moving, indicating ongoing insert-and-prune churn.

Privacy note

I am intentionally not pasting raw feedback_log_body contents because they may contain private conversation, local path, tool, or response data.

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

Windows Codex app continuously writes high-volume TRACE websocket logs to ~/.codex/logs_2.sqlite despite RUST_LOG=warn #29463
Codex SQLite feedback logs can write ~640 TB/year and rapidly consume SSD endurance #28224

Powered by Codex Action

Additional macOS reproduction on the current Codex app build.

Environment:

Platform: macOS, Apple internal SSD (APPLE SSD AP0512Z), SMART status Verified
Codex app: 26.616.71553, build 4265
Embedded CLI: codex-cli 0.142.0
App-server process: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
Active app-server PID during sample: 34793
RUST_LOG=info was set via launchctl getenv RUST_LOG, but persisted SQLite rows still included TRACE and DEBUG.
No SQLite trigger or DB mutation was applied during this sample; the checks below were read-only.

Files observed:

~/.codex/logs_2.sqlite      302,129,152 bytes (~288M)
~/.codex/logs_2.sqlite-wal   13,113,992 bytes (~13M)
~/.codex/logs_2.sqlite-shm       32,768 bytes

lsof showed the active Codex app-server process holding open handles to the DB, WAL, and SHM files.

60-second quiet sample:

Sample start: 2026-06-23 10:40:06 CST
Sample end:   2026-06-23 10:41:07 CST
max(id):      19457302 -> 19458968  (+1666)
count(*):     29226 -> 29235        (+9)
max(ts):      1782182405 -> 1782182465
WAL mtime:    kept advancing throughout the sample

This again suggests insert-and-prune churn: retained row count barely changed, but max(id) advanced by 1,666 in about a minute.

Rows visible in that sample window included:

TRACE log                                 870 rows, 473846 estimated bytes
TRACE codex_app_server::message_processor 13 rows
TRACE codex_api::sse::responses           11 rows
TRACE hyper_util::client::legacy::pool    11 rows
TRACE codex_app_server::outgoing_message   9 rows
TRACE codex_mcp::connection_manager        8 rows

Total visible TRACE rows in the sample window: 940.

The latest TRACE target=log samples were websocket/tungstenite noise, including:

WouldBlock
tokio-tungstenite ... Read.with_context read -> poll_read
WebSocketStream.with_context
Stream.poll_next
Received message ...
decompressing ... bytes in final frame
received frame <FRAME> ...

Current visible level distribution at collection time:

TRACE  15498 rows  53.03%
INFO   12331 rows  42.19%
DEBUG   1321 rows   4.52%
WARN      71 rows
ERROR      5 rows

Impact: this creates sustained writes to the persistent SQLite log sink during normal Codex app use. RUST_LOG=info did not stop these persisted TRACE rows, so the user-level mitigation does not appear to work for this path.

I have a private evidence package with raw command output, SQLite read-only queries, file metadata, hashes, and the 60-second WAL/SQLite sample. I am not attaching raw feedback_log_body contents publicly because they may include local paths, conversation IDs, command fragments, or response data.

Please fix this as soon as possible! This issue is widespread and serious. Thank you.

+1 Please fix this as soon as possible! This issue is widespread and serious. Thank you.

Additional current-build datapoint from macOS Desktop 26.616.71553 / build 4265, bundled/global codex-cli 0.142.0.

During active Desktop use, I temporarily removed a local SQLite mitigation trigger, recorded max(id)=707460512, waited 15 seconds, then restored the trigger. New retained rows over that window:

total_new_rows=1196
log|TRACE|934
codex_api::sse::responses|TRACE|138
codex_app_server::outgoing_message|TRACE|37
codex_mcp::connection_manager|TRACE|20
codex_config::loader::macos|DEBUG|8
codex_core::stream_events_utils|DEBUG|7
feedback_tags|INFO|7
codex_models_manager::cache|INFO|6
codex_models_manager::manager|INFO|5
codex_app_server::message_processor|TRACE|4
codex_config::loader::layer_io|DEBUG|4
codex_core::session::turn|TRACE|4
codex_core::stream_events_utils|INFO|4
codex_api::endpoint::responses_websocket|TRACE|3
codex_core::spawn|TRACE|3

Rows matching the previously noisy targets (log, codex_otel.log_only, codex_otel.trace_safe, codex_api::endpoint::responses_websocket, codex_api::sse::responses) were 1075 in the same 15-second window.

This is another confirmation that the current Desktop app-server path is still persisting high-volume TRACE target=log rows on the shipped 0.142.0 build. The local mitigation trigger was restored after the measurement.

Additional macOS data point from another machine: I can still reproduce persistent SQLite TRACE churn after restarting Codex Desktop and after manually compacting the log DB.

Environment

Platform: macOS arm64
Codex Desktop: 26.616.71553, build 4265
Embedded CLI: codex-cli 0.142.0
PATH CLI: codex-cli 0.142.0
Active Desktop app-server process:
- /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
A separate older stdio app-server process was also present, but recent rows were dominated by the restarted Desktop app-server process.
RUST_LOG observed in the process environment: warn
~/.codex/logs_2.sqlite journal mode observed earlier: wal

I am not attaching or pasting raw feedback_log_body rows because they can contain private conversation/tool/local-path details.

Before manual cleanup

Before compacting, the local log files were:

~/.codex/logs_2.sqlite       703M
~/.codex/logs_2.sqlite-wal   128M
~/.codex/logs_2.sqlite-shm   256K

The table showed high churn/free space:

rows:           143001
page_count:     179883
freelist_count: 133819
page_size:      4096

A 30 second sample after restarting Codex Desktop showed continued insert-and-prune activity:

max(id) delta: about +1414 in 30s
approx rate:   45-50 inserted ids/sec

Recent 2 minute target distribution during that sample was still dominated by TRACE rows:

TRACE log                                      1071 rows
TRACE codex_mcp::connection_manager             55 rows
TRACE codex_api::sse::responses                 51 rows
TRACE codex_app_server::outgoing_message        51 rows
TRACE codex_api::endpoint::responses_websocket   4 rows, about 2.17 MiB estimated

After manual cleanup / compaction

I then retained only a small recent window of diagnostic rows and ran checkpoint/VACUUM. Integrity check returned ok, and the files were reduced to:

15:53:24 local
logs_2.sqlite       1.8M
logs_2.sqlite-wal     0B
logs_2.sqlite-shm   256K
rows:               1855

However, without changing anything else, Codex Desktop quickly started growing the files again. About 80 seconds later:

15:54:44 local
logs_2.sqlite       38M
logs_2.sqlite-wal   37M
logs_2.sqlite-shm   256K
rows:               2061
max(id):            243430233

Rows retained in the last 5 minutes at that point:

TRACE log                                      1054 rows, 0.18 MiB estimated
TRACE codex_api::sse::responses                621 rows, 0.60 MiB estimated
TRACE codex_mcp::connection_manager            187 rows, 0.19 MiB estimated
TRACE codex_app_server::outgoing_message        48 rows
DEBUG codex_core::stream_events_utils           45 rows
TRACE codex_api::endpoint::responses_websocket   8 rows, 0.57 MiB estimated

Expected

After rust-v0.142.0 / #29432 / #29457, the persistent SQLite sink should not continue high-frequency insert-and-prune churn dominated by TRACE target=log and related websocket/SSE targets during normal Desktop use.

Actual

Even after restart and after compacting the database down to a tiny valid state, the active Desktop app-server began writing again immediately. The on-disk DB/WAL grew from about 1.8M + 0B to about 38M + 37M within roughly 80 seconds, and the retained recent rows were still dominated by TRACE targets.

This makes the current behavior look like an incomplete fix or a remaining Desktop-app persistent-log sink path that still persists noisy TRACE rows.

I’m seeing what looks like the same underlying SQLite churn pattern on macOS, though my setup differs slightly: I have not specifically disabled analytics/plugins, and the running process is codex app-server --analytics-default-enabled.

I posted the fuller diagnostic here: #29612 (comment)

Summary of my repro:

Codex Desktop: 26.616.71553 • Released Jun 22, 2026
Bundled CLI: codex-cli 0.142.0
Platform: macOS 15.3, Darwin 24.3.0, arm64
Launch method: macOS GUI / Spotlight
Running process: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
Actual process env includes: RUST_LOG=warn

Over a 5-minute idle sample, with Codex Desktop open in the background but not actively used:

new_insert_ids=2975
retained_row_delta=0
insert_ids_per_sec=9.92

The DB was healthy and compacted beforehand, so this did not appear to be leftover pre-fix bloat:

quick_check=ok
integrity_check=ok
journal_mode=wal
freelist_count=115
logs_2.sqlite≈63M
logs_2.sqlite-wal≈4.1M

The exact sampled ID window showed retained rows dominated by TRACE target=log:

TRACE log                                998 rows
TRACE hyper_util::client::legacy::pool     2 rows

All retained rows from that sampled window belonged to the same app-server process.

So this looks like the same post-0.142.0 behavior: file size may remain bounded, but the app-server still continuously inserts and prunes persistent SQLite log rows while idle, mostly TRACE target=log, despite RUST_LOG=warn.

Additional macOS datapoint after cleanup/compaction, from another affected machine.

I can still reproduce the post-0.142.0 insert-and-prune churn after fully compacting the log DB. The historical bloat was cleaned up successfully, but the current app-server continues to persist TRACE target=log rows.

Environment

Platform: macOS arm64
Codex Desktop: 26.616.71553
CFBundleVersion: 4265
CLI: codex-cli 0.142.0
Running process: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
App-server PID/start: 9920, Wed Jun 24 10:19:03 2026 AEST
Process env: RUST_LOG=warn
Active DB: ~/.codex/logs_2.sqlite

The active app-server process was holding open handles to logs_2.sqlite, logs_2.sqlite-wal, and logs_2.sqlite-shm.

Before cleanup

Before manually clearing/compacting the DB, the files had grown to:

~/.codex/logs_2.sqlite      811M
~/.codex/logs_2.sqlite-wal   68M
~/.codex/logs_2.sqlite-shm  160K

SQLite stats at that point showed this was mostly free pages from churn:

rows:           24495
page_size:       4096
page_count:    207595
freelist_count:195037
estimated_payload_mb: 26.66

After DELETE FROM logs, resetting the sequence, and VACUUM, the DB compacted successfully:

~/.codex/logs_2.sqlite      5.3M
~/.codex/logs_2.sqlite-wal  4.0M-4.3M
~/.codex/logs_2.sqlite-shm   32K

Repro after cleanup / restart

Immediately after cleanup and restart, rows began accumulating again, dominated by TRACE target=log:

rows: 1062
seq: 4027
TRACE log: 978 rows

A 20-second sample shortly after restart showed much higher active churn:

window_start=2026-06-24 10:19:48 AEST
window_end=2026-06-24 10:20:08 AEST
delta_seq=15923
rows_per_sec=796.1
delta_count=4

Top retained rows from that ID window:

TRACE log                                      932 rows, 0.176 MiB
TRACE codex_app_server::outgoing_message       39 rows
TRACE hyper_util::client::legacy::pool          7 rows
TRACE hyper_util::client::legacy::client        6 rows
TRACE codex_api::sse::responses                 2 rows

A later quieter 30-second sample still showed the same insert-and-prune pattern with file sizes bounded:

window_start=2026-06-24 10:23:52 AEST
window_end=2026-06-24 10:24:22 AEST
start_seq=67982
end_seq=68515
delta_seq=533
rows_per_sec=17.8
start_count=2049
end_count=2049
delta_count=0
start_db=5541888
end_db=5541888
delta_db=0
start_wal=4486712
end_wal=4486712
delta_wal=0

Top retained rows from that sampled ID window:

TRACE log                                      475 rows, 0.212 MiB
TRACE codex_app_server::outgoing_message       12 rows
TRACE hyper_util::client::legacy::pool         10 rows
TRACE hyper_util::client::legacy::client        6 rows
TRACE codex_api::sse::responses                 4 rows
DEBUG hyper_util::client::legacy::connect::http 4 rows

I am intentionally not pasting raw feedback_log_body values because they may contain private local paths, conversation, tool, or response data.

User-visible impact

Before cleanup, while several Codex tasks were open at the same time, I observed frequent Codex restarts and UI lag. Cleaning/VACUUM reduced the historical DB size from ~811M to ~5.3M, but the high-frequency TRACE target=log insert/prune behavior is still present on the current Desktop app-server.

So the remaining issue seems to be: 0.142.0 reduces/bounds some file-size growth, but the shipped macOS Desktop app-server still persists high-frequency TRACE target=log rows and continues SQLite churn even with RUST_LOG=warn.

Additional datapoint after updating to the newer macOS app build 26.616.81150 / build 4306.

Environment:

Codex app: 26.616.81150, build 4306
Embedded CLI: codex-cli 0.142.0
Active app-server process started after update: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
App-server PID during sample: 88553, started Wed Jun 24 09:20:21 2026
A local SQLite mitigation trigger existed before the test, so I temporarily removed it for a controlled 30-second sample, then restored it immediately afterward.

Controlled sample with the trigger disabled:

Sample window: 2026-06-24 09:22:34 CST to 09:23:05 CST
max(id):       19975524 -> 19975884 (+360)
count(*):      33614 -> 33965 (+351)
WAL size:      0 -> 1,030,032 bytes during the sample

Rows visible in the sample window:

TRACE log                                      441 rows, 271850 estimated bytes
TRACE codex_app_server::outgoing_message        4 rows
TRACE codex_mcp::connection_manager             4 rows
TRACE codex_api::endpoint::responses_websocket  1 row
TRACE codex_api::sse::responses                 1 row
TRACE codex_core::session::turn                 1 row
TRACE hyper_util::client::legacy::pool          1 row

Total TRACE rows in the 30-second window: 453.

The latest TRACE target=log rows are still websocket/tungstenite noise, including WouldBlock, tokio-tungstenite, WebSocketStream.with_context, Stream.poll_next, Received message, decompression, and <FRAME> parsing lines.

So this still reproduces on app build 4306; it does not appear fully fixed by this build.

现在macos的codex桌面端更新到了26.622.11653，还有这个问题吗

Additional macOS data point from Codex Desktop showing that the issue still reproduces on my machine, and that sqlite_home is currently the only effective local mitigation I found.

Environment

Platform: macOS, Apple SSD (diskutil: SMART Status Verified, TRIM supported)
Codex Desktop app-server command:
- /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
Codex app version exposed in local config/env:
- BROWSER_USE_CODEX_APP_VERSION=26.616.81150
Embedded CLI:
- codex-cli 0.142.0
SQLite DBs involved:
- ~/.codex/logs_2.sqlite
- ~/.codex/logs_2.sqlite-wal
- ~/.codex/state_5.sqlite-wal

Config tried

~/.codex/config.toml included:

[analytics]
enabled = false

[otel]
log_user_prompt = false
exporter = "none"
trace_exporter = "none"
metrics_exporter = "none"

The restarted app-server process also inherited:

RUST_LOG=warn

Confirmed via:

ps eww -p <app-server-pid> -o command= | tr ' ' '\n' | rg '^(RUST_LOG|CODEX|OTEL|RUST)='

Observed behavior before mitigation

After restarting Codex Desktop with the config above, a 60-second proc_pid_rusage sample still showed:

Codex-related processes total: 17.594 MB written / 60s
app-server alone:              17.082 MB written / 60s

The new app-server process had already accumulated substantial writes shortly after restart:

app-server start_written_mb=167.871

In an earlier active diagnostic window, app-server lifetime written bytes reached about:

app-server end_written_mb ~= 1373 MB

Recent rows in logs_2.sqlite still included persisted TRACE records despite RUST_LOG=warn:

TRACE  2957 rows  1.65 MB
DEBUG    95 rows  0.25 MB
INFO     84 rows  0.07 MB
WARN     13 rows  0.01 MB

Largest recent targets included:

codex_api::sse::responses                 TRACE
codex_api::endpoint::responses_websocket  TRACE
log                                       TRACE
codex_app_server::outgoing_message        TRACE
codex_mcp::connection_manager             TRACE

This confirms the SQLite feedback/log sink is still persisting TRACE-level data even when the process environment has RUST_LOG=warn and analytics/OTel exporters are disabled.

Why file size alone hides the issue

The retained DB size was not enough to estimate real write volume. The file stayed bounded while logs_2.sqlite-wal and SQLite insert/prune churn continued. This matches the pattern described in this issue: current file size can look moderate while SSD writes continue due to insert, WAL write, prune, checkpoint, and rewrite activity.

Local mitigation tested

I moved only Codex's SQLite-backed runtime state to a RAM disk:

sqlite_home = "/Volumes/CodexSQLiteRAM"

After restarting Codex Desktop, lsof confirmed these were all on the RAM disk:

/Volumes/CodexSQLiteRAM/logs_2.sqlite
/Volumes/CodexSQLiteRAM/logs_2.sqlite-wal
/Volumes/CodexSQLiteRAM/state_5.sqlite-wal
/Volumes/CodexSQLiteRAM/memories_1.sqlite-wal
/Volumes/CodexSQLiteRAM/goals_1.sqlite-wal

A follow-up 60-second sample still showed process-level writes:

Codex-related processes total: 8.230 MB written / 60s
app-server alone:              7.191 MB written / 60s

But visible internal SSD growth for the Codex SQLite files dropped to zero because the SQLite writes were on the RAM disk:

RAM_SQLITE_FILE_GROWTH_MB       0.118
SSD_OLD_SQLITE_FILE_GROWTH_MB   0.000
SSD_APP_LOG_FILE_GROWTH_MB      0.000
SSD_SESSION_FILE_GROWTH_MB      0.002

A broader 30-second scan of likely Codex internal-SSD paths showed only tiny non-SQLite growth:

~/.codex                                      0.000 MB
~/Library/Application Support/Codex           0.007 MB
~/Library/Caches/com.openai.codex             0.000 MB
~/Library/HTTPStorages/com.openai.codex       0.000 MB
~/Library/Logs/com.openai.codex               0.000 MB
TOTAL_BROAD_FILE_GROWTH_MB                    0.007 MB / 30s

So sqlite_home to RAM disk mitigates SSD wear, but it does not fix the root cause. It only moves the high-frequency SQLite WAL writes away from the internal SSD.

Expected fix

Please make the persistent SQLite log sink honor the same filtering as RUST_LOG / runtime log level, or provide a documented config option to disable persistent SQLite feedback logs or cap them aggressively. With RUST_LOG=warn, [analytics].enabled=false, and all [otel] exporters set to none, TRACE websocket/SSE/internal target=log records should not continue to be persisted to logs_2.sqlite during normal desktop usage.

Privacy note

I am not attaching the SQLite DB or raw feedback_log_body rows because they may contain private local paths, tool output, and conversation/session details.

Additional macOS datapoint after updating to Codex app 26.623.101652 / build 4674.

Environment:

Codex app: 26.623.101652, build 4674
Embedded CLI: codex-cli 0.142.5
Active app-server command: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
RUST_LOG was unset in the launchd environment at the time of the check.
A local SQLite mitigation trigger existed before the test, so I temporarily removed it for a controlled 30-second sample, then restored it immediately afterward and truncated the WAL again.
No raw feedback_log_body values are included here because they may contain private local paths, tool output, or conversation/session details.

Notable process state after a full app restart/update:

Two codex app-server processes were running and both held open handles to logs_2.sqlite, logs_2.sqlite-wal, and logs_2.sqlite-shm.

PID    PPID  started
94151  1     Fri Jul  3 12:07:17 2026
94919  94887 Fri Jul  3 12:07:36 2026

Controlled sample with the trigger disabled:

Sample window: 2026-07-03 12:09:43 CST to 12:10:13 CST
max(id):       20053926 -> 20054522 (+596)
count(*):      34196 -> 34792 (+596)
WAL size:      0 -> 1,833,432 bytes during the sample

Rows visible in the sample window:

level  target                                                            rows  estimated_bytes
TRACE  log                                                               577   175550
DEBUG  log                                                               8     419
TRACE  codex_api::sse::responses                                         3     3212
TRACE  hyper_util::client::legacy::pool                                  3     665
INFO   codex_app_server_transport::transport::remote_control::websocket  1     624
INFO   codex_core::stream_events_utils                                   1     4630
INFO   feedback_tags                                                     1     133
TRACE  codex_app_server::outgoing_message                                1     160
WARN   codex_app_server_transport::transport::remote_control::websocket  1     935

Rows by process UUID in the same sample window:

process_uuid                                    rows  trace_rows  debug_rows
pid:94151:1af60d6a-3dc5-4608-bc97-d376c47db0ba  315   315         0
pid:94919:1b9d9fd8-2306-4905-b74d-829cfe067eab  281   269         8

After restoring the local mitigation trigger and running PRAGMA wal_checkpoint(TRUNCATE), the local state was:

trigger_present: block_trace_debug_logs
logs_2.sqlite-wal size: 0

A follow-up 30-second observation with the trigger enabled persisted only INFO/WARN rows, while TRACE/DEBUG rows were not persisted by the local mitigation.

This suggests 0.142.5 / build 4674 may have improved the large websocket request-payload trace symptom from #30771, but the broader persistent TRACE/DEBUG SQLite churn still reproduces when the local SQLite mitigation is removed. In this run, both active app-server processes contributed rows.

Follow-up from the same affected Mac after updating to a newer Desktop build again.

Environment:

Platform: macOS 27.0 (26A5368g), arm64
Codex app: 26.623.101652, build 4674
Embedded CLI: codex-cli 0.142.5
PATH CLI: codex-cli 0.142.5
Active desktop app-server: /Applications/Codex.app/Contents/Resources/codex app-server --analytics-default-enabled
Active desktop app-server PID during check: 98496, started Mon Jul 6 10:49:08 2026
launchctl getenv RUST_LOG returned empty/unset during this check
A separate stdio app-server was also present: /Users/ron/.local/bin/codex app-server --listen stdio://, PID 3728
Local SQLite mitigation triggers were still enabled: sample_log_inserts, trim_sampled_logs

Important note: this is another protected-state check, not a raw insert-rate measurement. I did not remove the local mitigation this time. On this machine those triggers retain at most about one row every 10 seconds and keep only the newest 500 retained rows.

Current retained SQLite state after the sample:

PRAGMA quick_check: ok
journal_mode: wal
row_count: 500
trace_rows: 500
page_count: 32769
freelist_count: 32671
page_size: 4096
latest retained row: 2026-07-06 11:12:07 Asia/Shanghai

30-second protected sample:

before: count=500 max_id=189220 trace_rows=500 max_trace_id=189220 max_ts=1783307497 latest=2026-07-06 11:11:37
 after: count=500 max_id=189223 trace_rows=500 max_trace_id=189223 max_ts=1783307527 latest=2026-07-06 11:12:07

File sizes at the end of the check:

~/.codex/logs_2.sqlite      134221824 bytes
~/.codex/logs_2.sqlite-wal    4185952 bytes
~/.codex/logs_2.sqlite-shm      32768 bytes

Top retained targets after the sample:

TRACE log                                  480
TRACE hyper_util::client::legacy::pool      15
TRACE hyper_util::client::legacy::client     4
TRACE codex_app_server::outgoing_message     1

The newest retained rows immediately after the sample were still from the active Desktop app-server process UUID:

189225  2026-07-06 11:12:27  TRACE  log  pid:98496:06d227f2-204d-4c82-a197-bd163eb74d1f
189224  2026-07-06 11:12:17  TRACE  log  pid:98496:06d227f2-204d-4c82-a197-bd163eb74d1f
189223  2026-07-06 11:12:07  TRACE  log  pid:98496:06d227f2-204d-4c82-a197-bd163eb74d1f
189222  2026-07-06 11:11:57  TRACE  log  pid:98496:06d227f2-204d-4c82-a197-bd163eb74d1f

Conclusion from this machine: 26.623.101652 / build 4674 / codex-cli 0.142.5 still does not look fully fixed. The large websocket request-payload trace symptom may be improved, as others noted for #30771, but the broader persistent SQLite TRACE sink still appears active: retained rows remain 500/500 TRACE and max_trace_id continues to advance under the local sampled-retention trigger.

I am not pasting raw feedback_log_body values because they may include local paths, tool output, or conversation/session details.

Additional reproduction on ChatGPT-integrated Codex / 0.144.0-alpha.4

I can still reproduce the bridged TRACE target=log persistence on a newer macOS build that embeds Codex inside the ChatGPT app.

Environment

macOS, Apple Silicon
ChatGPT app: 26.707.31428 (build 5059)
Embedded binary: /Applications/ChatGPT.app/Contents/Resources/codex
Embedded CLI version: codex-cli 0.144.0-alpha.4
SQLite DB: ~/.codex/logs_2.sqlite

Evidence collected before installing any local workaround

The current app-server process persisted 436 rows with target=log between:

2026-07-10 08:27:01 +0800
2026-07-10 08:27:05 +0800

Example sanitized messages included:

TRACE log: deregistering event source from poller
TRACE log: shouldn't retry!

At the time of inspection, the retained target counts included:

log                              2333
codex_api::sse::responses        1654
hyper_util::client::legacy::pool  547

No retained rows were found for codex_otel.log_only or codex_otel.trace_safe, so that part of the filtering appears effective. However, bridged target=log events are still reaching SQLite, despite #29599 being intended to reject them inside the sink.

This appears to be either:

a regression after Stop persisting bridged log events #29599 / 0.143.0, or
a different logging path in the ChatGPT-integrated app build that bypasses the sink-level metadata.target() == "log" guard.

Local confirmation/workaround

After the measurement, I installed a SQLite BEFORE INSERT ... RAISE(IGNORE) trigger for the noisy targets. A test insert was ignored, and normal non-blocked log IDs continued advancing while the blocked-target maximum ID stayed unchanged. This is only a local workaround; filtering before formatting/queueing remains preferable.

Raw payloads, thread IDs, local paths, and conversation contents are intentionally omitted.

Follow-up from the same affected Mac after updating to the newer ChatGPT-integrated Codex build.

Environment:

Platform: macOS 27.0 (26A5378j), arm64
ChatGPT app: 26.707.61608, build 5200
Embedded binary: /Applications/ChatGPT.app/Contents/Resources/codex
Embedded CLI version: codex-cli 0.144.0-alpha.4
command -v codex now resolves to the ChatGPT embedded binary on this machine
The previous standalone /Applications/Codex.app/Contents/Resources/codex binary was not present during this check
Active ChatGPT-integrated app-server: /Applications/ChatGPT.app/Contents/Resources/codex -c features.code_mode_host=true app-server --analytics-default-enabled
Active ChatGPT-integrated app-server PID during check: 92069, started Mon Jul 13 10:03:30 2026
A separate stdio app-server was also present: ~/.local/bin/codex app-server --listen stdio://, PID 89531
launchctl getenv RUST_LOG returned empty/unset during this check
Local SQLite mitigation triggers were still enabled: sample_log_inserts, trim_sampled_logs

Important note: this is another protected-state check, not a raw insert-rate measurement. I did not remove the local mitigation this time. On this machine those triggers retain at most about one row every 10 seconds and keep only the newest 500 retained rows.

Current retained SQLite state after the sample:

PRAGMA quick_check: ok
journal_mode: wal
row_count: 500
trace_rows: 451
debug_rows: 33
info_rows: 10
warn_rows: 6
error_rows: 0
page_count: 32769
freelist_count: 32649
page_size: 4096
latest retained row: 2026-07-13 10:33:48 Asia/Shanghai

30-second protected sample:

before: count=500 max_id=221476 trace_rows=452 max_trace_id=221475 debug_rows=32 info_rows=10 warn_rows=6 error_rows=0 latest=2026-07-13 10:33:18
 after: count=500 max_id=221479 trace_rows=451 max_trace_id=221479 debug_rows=33 info_rows=10 warn_rows=6 error_rows=0 latest=2026-07-13 10:33:48

File sizes at the end of the check:

~/.codex/logs_2.sqlite       134221824 bytes
~/.codex/logs_2.sqlite-wal    44388912 bytes
~/.codex/logs_2.sqlite-shm       32768 bytes

Top retained targets after the sample:

TRACE  log                                   333
TRACE  hyper_util::client::legacy::pool       27
TRACE  codex_app_server::outgoing_message     25
TRACE  hyper_util::client::legacy::client     25
TRACE  codex_api::sse::responses              24
TRACE  codex_app_server::message_processor    17
DEBUG  codex_http_client::default_client      15
DEBUG  codex_core::stream_events_utils        10
DEBUG  opentelemetry-otlp                      8
INFO   codex_http_client::custom_ca            6
WARN   codex_app_server_transport::transport::remote_control::websocket  5

The newest retained rows immediately after the sample included new TRACE rows from both the ChatGPT-integrated app-server and the stdio app-server:

221479  2026-07-13 10:33:48  TRACE  codex_app_server::message_processor  pid:89531:7334f365-0926-407f-aac9-abe8fc58abf9
221478  2026-07-13 10:33:38  TRACE  codex_api::sse::responses            pid:92069:71bd96dc-8afa-4d14-85b4-d909fc2345a0
221477  2026-07-13 10:33:28  DEBUG  codex_http_client::default_client    pid:92069:71bd96dc-8afa-4d14-85b4-d909fc2345a0
221476  2026-07-13 10:33:18  DEBUG  codex_core::stream_events_utils      pid:92069:71bd96dc-8afa-4d14-85b4-d909fc2345a0
221475  2026-07-13 10:33:08  TRACE  codex_app_server::outgoing_message   pid:92069:71bd96dc-8afa-4d14-85b4-d909fc2345a0

Conclusion from this machine: the old 500/500 TRACE protected-state symptom has changed in the ChatGPT-integrated 0.144.0-alpha.4 build: retained rows now include DEBUG/INFO/WARN as well. However, the broader issue still does not look fully fixed here. max_trace_id continued to advance during the protected sample, and TRACE target=log remains the largest retained target. This matches the concern in the newer ChatGPT-integrated reports: part of the older filtering appears improved, but bridged TRACE/log-style events still reach the persistent SQLite DB.

I am not pasting raw feedback_log_body values because they may include local paths, tool output, or conversation/session details.

@darlingm

Our series of fixes for this problem are in the latest shipping desktop app, CLI, and IDE extension.

Not quite yet. 0.144.1 does not have #31789, #31790, #31791, and #31792. 0.145.0-alpha.4 includes them, so 0.145.0 likely will too.

Also, please see #32496. A different root cause is causing about 1.3 TB/year in unnecessary writes, according to the numbers in the issue. (Much less than was on this issue.) Codex is continually rewriting models_cache.json and scope_v3.json.

Originally posted by @darlingm in #28224 (comment)

I’ve completed an independent reference implementation and validation effort for the persistent SQLite TRACE/DEBUG issue described here.

Repository:

https://github.com/Freyliu0516/Codex-Log-Doctor

Proposed upstream change

The upstream patch is deliberately small and separate from the user-side diagnostic tool:

change the persistent SQLite sink’s default level from TRACE to WARN;
keep the existing target-specific exclusions;
filter low-level events before they are formatted, queued, or sent to SQLite;
add a regression test that emits 100,000 TRACE events and 100,000 DEBUG events;
verify that SQLite retains only the expected WARN event.

Reference patch:

https://github.com/Freyliu0516/Codex-Log-Doctor/tree/main/upstream

The independent tool does not patch or replace an installed Codex binary. The upstream/ directory exists only as a reviewable reference implementation.

Validation

The patched Codex source was tested against the pinned upstream base:

codex-state primary tests: 151 passed;
binary tests: 3 passed;
doctests: 1 passed;
persistence regression:
- 100,000 TRACE iterations;
- 100,000 DEBUG iterations;
- 0 unexpected rows persisted;
- the expected WARN row was retained.

The separate Codex Log Doctor repository also passed:

formatting;
workspace-wide Clippy;
all workspace tests;
release builds on Linux, macOS, and Windows;
supply-chain checks;
source-install smoke tests;
upstream patch drift checks.

The repository additionally contains 18 deterministic component and failure-mode scenarios covering SQLite/WAL contention, queue saturation, backup and rollback, interrupted operations, unsupported schemas, and report redaction. These are supporting tests, not a claim that all 18 scenarios exercise the complete Codex app-server pipeline.

Public CI evidence is linked from the merged initial implementation PR:

Freyliu0516/Codex-Log-Doctor#1

User-side mitigation

The tool can install a fingerprint-gated SQLite BEFORE INSERT trigger that rejects TRACE rows. It requires:

an exact supported schema fingerprint;
no detected Codex process using the database;
exclusive SQLite access;
sufficient disk space;
a verified SQLite backup;
explicit --apply;
rollback metadata and post-operation integrity checks.

This is intended only as a reversible workaround. It reduces retained TRACE rows, MAX(id) growth, WAL writes, indexing, and subsequent pruning, but it does not eliminate upstream event construction, formatting, queueing, transactions, or attempted INSERT execution.

Early filtering in Codex remains the preferred fix.

Request

Would the Codex team be open to reviewing the persistence-filter approach?

If changing the persistent SQLite default from TRACE to WARN is consistent with the intended feedback and diagnostics behavior, I would appreciate guidance on whether the team would invite a focused upstream PR limited to:

codex-rs/state/src/log_db.rs;
the SQLite sink regression test;
any required documentation or configuration updates.

I’m also happy to adjust the proposed target-specific policy if there are operational events that the feedback pipeline must continue retaining below WARN.

I can still reproduce significant persistent SQLite TRACE churn on the current ChatGPT-integrated Codex desktop build, and I think this deserves clarification after #28224 was closed.

Environment

Platform: macOS / Apple Silicon
App: /Applications/ChatGPT.app
ChatGPT.app version: 26.707.72221
Build: 5307
Bundled Codex CLI: codex-cli 0.144.2
Primary DB: ~/.codex/logs_2.sqlite
Also observed project-scoped CODEX_HOME DBs such as client-hub-codex-home/logs_2.sqlite

Controlled test

I had been using a local SQLite stopgap trigger:

CREATE TRIGGER IF NOT EXISTS block_log_inserts
BEFORE INSERT ON logs
BEGIN
  SELECT RAISE(IGNORE);
END;

With the trigger enabled, short samples show:

id_delta=0
count_delta=0
WAL stable

I then temporarily removed the trigger for a controlled test and restored it immediately afterward.

Result: within roughly a one-minute active Codex window, the global ~/.codex/logs_2.sqlite accumulated about 2.6k rows and the WAL grew to about 4.15 MiB.

Dominant persisted targets were still TRACE/DEBUG-style internal plumbing:

TRACE codex_api::sse::responses
TRACE codex_mcp::connection_manager
TRACE codex_app_server::outgoing_message
TRACE hyper_util::client::legacy::*
DEBUG codex_core::stream_events_utils

After restoring the trigger, another short sample returned to:

id_delta=0
count_delta=0
WAL stable

Why I am posting

#28224 was closed as fixed, but the current shipping desktop app on my machine still bundles codex-cli 0.144.2. From the discussion around #28224, it sounds like the more relevant logging fixes (#31789, #31790, #31791, #31792) are expected in the 0.145.0-alpha.4+ line, not necessarily in this bundled desktop version.

So I would like to clarify:

Is this residual persistent TRACE/DEBUG SQLite churn considered expected behavior?
Which desktop app / bundled CLI version should include the full fix for Stop persisting RMCP service traces #31789-Summarize streamed response item logs #31792?
Is there an official threshold for acceptable idle/active Codex local disk writes?
Should users keep local mitigations like block_log_inserts until the bundled desktop CLI reaches the fixed line?
Can OpenAI add an I/O regression test for idle and active Codex usage, so this class of issue does not regress silently?

I would also encourage other affected users to share comparable measurements with:

app version/build
bundled CLI version
platform
sample duration
MAX(id) delta
retained row delta
WAL byte delta
top persisted log targets

@jif-oai

Fresh measurements on codex-cli 0.144.3 (Linux/ext4), plus a minimal tested patch and an A/B benchmark. TL;DR: most of the remaining churn is flush policy, not log content — batching the sink by bytes/severity instead of every 2 s cuts its physical writes ~5x on a steady-state workload, down to SQLite's WAL+checkpoint double-write floor, with zero change to what /feedback sees.

Measurements: the writes are flush overhead, not content

Four resumed TUI sessions left idle; write rates from /proc/<pid>/io deltas, content attribution from querying logs_2.sqlite (read-only) for rows inserted inside the same instrumented window:

metric	value
physical writes in a 59 s idle window (2 busiest processes)	28.2 MB
log rows retained in that same window	630 rows / 774 KB
physical-to-retained write amplification	~36x
combined idle write rate, 4 sessions	~353 KB/s ≈ 30 GB/day
one idle session running 9.3 days	123 GB lifetime writes

The DB itself stays at ~25 MB — retention caps work. The wear comes from the durability policy: LOG_FLUSH_INTERVAL = 2s means ~43,000 transactions/day/process, each paying WAL page images + (at retention-cap steady state, which every long-lived session reaches) the windowed prune DELETE inside the same transaction.

Row-level attribution also shows why target muting (#29457, #29599, #31789, #31791) can't converge: after accounting for those mutes, idle traffic is dominated by codex's own targets — codex_app_server::outgoing_message (TRACE), codex_api::sse::responses (TRACE), codex_core::stream_events_utils (DEBUG), feedback_tags (INFO). Muting those defeats /feedback; keeping them keeps the churn. The reasoning that closed #23275 ("bounding individual call sites is brittle and misses future tracing events") applies equally to per-target filters.

Patch: flush by bytes/severity/60 s instead of every 2 s

Branch: https://github.com/friday-james/codex-1/tree/fix/log-db-write-churn — 3 files, +127/−9, mostly tests.

Flush the sink buffer on any of: ~1 MiB of estimated pending content, a 60 s tick, or a WARN/ERROR entry (flushed immediately, so crash-relevant diagnostics still reach disk right away).
/feedback already calls log_db.flush().await before reading (feedback_processor.rs), so feedback content is byte-identical, even mid-session.
LogEntry::estimated_bytes() is factored out and shared by the sink buffer and the insert path, so the write-side threshold and partition budgets agree.
Acknowledged trade-off: on SIGKILL, up to 60 s of TRACE/DEBUG tail can be lost (WARN/ERROR are not affected, and any explicit flush path is not affected). Today's policy loses up to 2 s.

A/B benchmark (each phase an isolated process)

Harness: https://github.com/friday-james/codex-1/tree/bench/log-db-write-churn (state/examples/write_churn_bench.rs). It replays the measured idle profile — 5.3 rows/s, ~1.2 KiB/row, thread partition pre-filled past its retention cap (steady state of a long-lived session) — for 3 minutes per policy, measuring /proc/self/io write_bytes. Two runs, results identical to ±0.2%:

old policy (2 s / 128 rows):        14.5 MB in 180 s  ->  7.02 GB/day
new policy (60 s / 1 MiB / sev):     2.9 MB in 180 s  ->  1.42 GB/day   (~4.9x less)

The new policy's 2.9 MB against 1.17 MB of raw content is SQLite's WAL-then-checkpoint double write plus prune/B-tree overhead — i.e. the patch removes nearly all removable churn; going lower requires persisting less content or relaxing durability guarantees. Note the benchmark is single-process and therefore conservative: live sessions share one logs_2.sqlite across processes, where checkpoint contention pushes the old policy 2–3x higher than the bench (and in the limit produces the unbounded-WAL growth reported in #28997).

Happy to open this as a PR per the invitation policy in contributing.md, or the approach is free to take as-is. Two further options if there's interest: prune hysteresis (once a partition crosses its cap, delete down to ~75% so steady-state inserts don't re-run the windowed DELETE every batch), and honoring an env override for the sink filter, which several reports expected RUST_LOG to do (#29463, #31111).

cc @jif-oai @charley-oai

Request for clarification: current shipping build still lacks the broader SQLite fix

Following up with the current shipping macOS build:

ChatGPT/Codex app: 26.707.91948, build 5440
Bundled CLI: codex-cli 0.144.5
PATH CLI: codex-cli 0.144.3
npm stable: 0.144.5

The rust-v0.144.5 release contains only the dangerous-command detection fix in #33455. Its persistent log filter still uses:

Targets::new().with_default(LevelFilter::TRACE)

Source: https://github.com/openai/codex/blob/rust-v0.144.5/codex-rs/state/src/log_db.rs

The 0.144.4...0.144.5 release comparison also does not include #31789, #31790, #31791, or #31792. This appears inconsistent with the statement in #28224 that the series of fixes was already in the latest shipping desktop app, CLI, and IDE extension.

My local SQLite database is currently protected by this reversible stopgap:

CREATE TRIGGER block_log_inserts
BEFORE INSERT ON logs
BEGIN
  SELECT RAISE(IGNORE);
END;

Protected-state verification on the current build shows rows=0, seq=0, and max(id)=0. I did not remove the trigger for another raw sample because the shipping 0.144.5 source contains no relevant persistent-log change. I am not posting any raw feedback_log_body values because they may contain private paths, command fragments, tool output, or conversation context.

Could the maintainers please clarify:

Is LevelFilter::TRACE still the intended production default for the persistent SQLite feedback log?
Which exact stable CLI version and desktop build will contain the complete fix, including the relevant target/callsite reductions and any write-flush policy change?
Is there a planned ETA or release line for that fix?
Until that version ships, should affected users continue using a BEFORE INSERT ... RAISE(IGNORE) trigger?
If the trigger is not recommended, what is the officially supported mitigation that prevents excessive writes without losing essential /feedback diagnostics?
Can a regression test or documented acceptable idle/active write threshold be added so users can verify the fix?

A version-specific answer would help users know when it is safe to remove the trigger and restore normal persistent diagnostic logging.

Follow-up after updating this Mac to the current ChatGPT-integrated Codex build.

Environment now:

Platform: macOS 27.0 (26A5378n), Apple Silicon / arm64
App: /Applications/ChatGPT.app
ChatGPT app: 26.715.31925, build 5551
codex resolves to: /Applications/ChatGPT.app/Contents/Resources/codex
CLI: codex-cli 0.145.0-alpha.18
Old standalone /Applications/Codex.app is not present on this machine
RUST_LOG is unset

I kept the local SQLite mitigation in place for this check. The active triggers are still:

sample_log_inserts
trim_sampled_logs

So this is a protected-state verification only; I did not remove the mitigation for another raw write-churn sample, and I am not posting raw feedback_log_body values publicly.

Current protected-state sample from ~/.codex/logs_2.sqlite:

quick_check:     ok
journal_mode:    wal
rows:            500
TRACE rows:      104
DEBUG rows:      128
INFO rows:       259
WARN rows:       9
ERROR rows:      0
sqlite size:     134,221,824 bytes
wal size:        4,194,192 bytes
page_count:      32769
freelist_count:  32650
page_size:       4096

30-second retained-row sample with the mitigation still enabled:

max(id):        240689 -> 240692  (+3)
rows:           500 -> 500
TRACE rows:     105 -> 104
max TRACE id:   240689 -> 240689
sqlite size:    unchanged at 134,221,824 bytes
wal size:       unchanged at 4,194,192 bytes

Newest retained rows in that sample were DEBUG codex_http_client::default_client and INFO codex_http_client::custom_ca; the newest retained TRACE row stayed at id 240689 (TRACE codex_app_server::message_processor). The retained sample still includes TRACE entries from codex_app_server::message_processor, codex_api::sse::responses, and codex_app_server::outgoing_message, but I cannot infer the unmitigated raw insert rate from this protected sample.

Given #4988716077 and #4989141674, this updated app/CLI datapoint may help separate two questions:

whether the 0.145.x ChatGPT-integrated build has the target/callsite reductions expected from the 0.145 alpha line;
whether the broader SQLite sink write/flush policy issue remains separate from the target filtering work.

A version-specific maintainer answer would still be useful: is 26.715.31925 / build 5551 / codex-cli 0.145.0-alpha.18 expected to be safe enough for affected users to remove local SQLite mitigations, or should users keep them until a stable desktop/CLI release with an explicit persistent-log fix ships?

Hey, I had a look at this.

The SQLite sink already has a guard that drops bridged events with the log target before they reach the queue. The missing piece was a test that proves the guard still works when the outer tracing filter is permissive, which is the path that allowed this noise through in the first place.

I've added that regression test. It sends a bridged target="log" event alongside a normal Codex state event, flushes the sink, and checks that only the normal event was written to SQLite.

Branch: https://github.com/TheJhyeFactor/codex/tree/fix/sqlite-log-bridge-regression-29532

I ran CARGO_NET_OFFLINE=true cargo test -p codex-state locally on macOS. All 153 unit tests, 3 binary tests, and 1 doctest passed.

I couldn't open the PR because GitHub returned a CreatePullRequest permission error for my account, so I've linked the tested branch here for review.

Follow-up after another ChatGPT-integrated Codex update on the same Mac.

Environment now:

ChatGPT app: 26.715.72359, build 5718
Embedded CLI: codex-cli 0.145.0-alpha.30
codex still resolves through /Applications/ChatGPT.app/Contents/Resources/codex
The old standalone /Applications/Codex.app is still not present
RUST_LOG is unset

I kept the local SQLite mitigation enabled for this check, so this is still only a protected-state verification. I did not remove the mitigation for an unbounded raw write-churn sample, and I am not posting raw feedback_log_body values or detailed local SQLite metadata publicly.

Qualitatively, in the protected sample I no longer see bridged TRACE target=log among the newest/top retained entries, which seems consistent with the regression-test discussion in #5018554274. However, normal Codex TRACE targets are still present in the retained sample, and this protected check cannot answer the broader SQLite sink write/flush-policy question raised in #4988716077.

Could maintainers clarify two version-specific points for affected users?

Is 26.715.72359 / build 5718 / codex-cli 0.145.0-alpha.30 expected to contain the relevant macOS persistent-log target filtering fix?
Separately, is there any shipped or planned fix for the SQLite sink write/flush-policy churn, or should users keep local SQLite mitigations until a stable release explicitly mentions that broader fix?

Analysis (community)

Thanks for tracking this — attempting a residual-focused analysis aligned with docs/contributing.md (invitation-only: analysis in-thread, no unsolicited PR).

Issue under review: macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0 #29532 — macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0
Theme / radar cluster: residual TRACE write after claimed fix (logs_2.sqlite SSD)
Reporter posture: static + local evidence where available; not claiming a maintainer invite.

Observed (from issue + public residual signal)

Summary After upgrading to `rust-v0.142.0`, I can still reproduce persistent SQLite log churn on macOS in `~/.codex/logs_2.sqlite`. This looks like a partial fix: - `#29432` appears to have helped: `codex_api::endpoint::responses_websocket` is much lower than before. - However, `#29457` does not appear to fully solve the persistent SQLite churn in this app build/runtime. The release note says noisy persistent log targets were filtered, and the PR says bridged `target=log` events should be exc

Local residual measurement (2026-07-24, this reporter)

On a host running codex-cli 0.146.0-alpha.6 (newer than 0.142–0.145 residual reports):

Signal-14 INTERIM evidence (from qwen_s14 mid-flight)

captured: 2026-07-24T15:10:05Z

/Users/bing/.codex/logs_2.sqlite size=1778548736 mtime=Fri Jul 24 23:09:38 2026
/Users/bing/.codex/logs_2.sqlite-wal size=30405632 mtime=Fri Jul 24 23:10:04 2026
codex version: codex-cli 0.146.0-alpha.6

level counts (ro)

TRACE|94794
DEBUG|21528
INFO|19618
WARN|796
ERROR|370

rows/max_ts: 137106|2026-07-24 23:10:05

This is not a claim that every install still burns TB/year — it is a version-pinned observation that TRACE still lands in the SQLite sink after the #28224-era mitigations on a current alpha, consistent with residual issues still open.

Root-cause hypothesis (to be confirmed on `main`)

Feedback/log pipeline still admits TRACE (or default level too verbose) into logs_2.sqlite.
WAL/checkpoint policy does not cap growth under idle or streaming load.
Partial cut of TRACE paths may leave high-volume targets writing.

High-level fix outline (not a PR)

Default sink level: never store TRACE in SQLite unless explicitly enabled.
Honor RUST_LOG / app log level end-to-end into the SQLite writer.
Hard caps: max DB size / WAL size / rotation with user-visible warning.
Fail-first tests: write-rate budget under idle + streaming fixtures.

Tests

Unit: writer rejects TRACE when configured level is INFO.
Integration: after N seconds idle with default config, row growth rate < bound.

Questions for maintainers

Is the intended long-term policy documented for this area (especially sandbox gitdir / log TRACE defaults)?
Preferred residual anchor issue if several open threads describe the same mechanism?
If the high-level outline above is roughly aligned, would an invited PR with failing-first tests be welcome later — or is an internal fix preferred?

Happy to refine with more measurements or code pointers. No unsolicited PR from me.

Analysis (community)

Residual analysis for #29532 (TRACE churn remains after rust-v0.142+ claims) — invitation-only under docs/contributing.md. No unsolicited PR.

Distinct from #17320 (default TRACE policy) and #28997 (WAL budget): this thread is about fix claims vs still-shipping residual on current builds.

Deterministic evidence (code + runtime)

Code (5dd992acd3):

codex-rs/state/src/log_db.rs default_filter():
- .with_default(LevelFilter::TRACE) (L55)
- only a short denylist of noisy targets (hyper_util→WARN, log→OFF, a few codex_otel* / codex_api::responses_websocket_timing OFF, rmcp::service→INFO)
High-fanout product targets are not demoted: SSE responses, app-server outgoing messages, TUI markdown stream still admit TRACE into the SQLite sink via LogDbLayer::on_event.
Unit tests in codex-rs/state/src/log_db_filter_tests.rs still document TRACE retention for non-filtered targets.

Runtime (codex-cli 0.146.0-alpha.6):

logs_2.sqlite ≈ 1778548736 bytes; TRACE rows 94907 of ≈137150 total.
Top TRACE targets on this host:
- codex_api::sse::responses → 37252
- codex_app_server::outgoing_message → 19558
- codex_tui::markdown_stream → 15085
That distribution shows residual TRACE is product-path, not only third-party crates already WARN-filtered.

Root-cause hypothesis

Partial target denylists / “we fixed TRACE spam” claims did not change the default LevelFilter::TRACE for the SQLite sink, nor demote the highest-volume first-party streaming targets. Users on 0.146.x still observe multi-GB DBs and TRACE-dominant histograms.

High-level fix outline (not a PR)

Ship default sink level INFO (or DEBUG) with explicit TRACE opt-in.
Demote first-party high-fanout targets (codex_api::sse::responses, codex_tui::markdown_stream, codex_app_server::outgoing_message) even if global default stays DEBUG.
Release note matrix: version → default sink level → measured TRACE share under a 5-minute stream repro.
Fail-first: assert default filter does not retain generic TRACE for those three targets.

Questions for maintainers

Which release was supposed to close the residual claimed on macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0 #29532, and against which default?
Preferred anchor among macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0 #29532 / Excessive SQLite WAL writes during streaming due to TRACE logs ignoring RUST_LOG #17320 / logs_2.sqlite-wal grows without bound into tens of GB #28997 for the “still TRACE by default on 0.146.x” proof?
Would an invited PR for default-level + first-party demotions with tests be useful?

No unsolicited PR from me.

Additional current-stable reproduction and bounded mitigation result (2026-08-15):

Environment: macOS 26.6.1 (arm64), official stable codex-cli 0.147.0, persistent app-server.
A new sanitized macOS Disk Writes diagnostic for that exact 0.147.0 process reports 8,589.94 MB over 25,456 seconds (337.44 KB/s versus a 99.42 KB/s threshold), with representative stacks dominated by SQLite pwrite, WAL checkpoint, and fsync. A second current-version report records 2,147.49 MB over 44,782 seconds (47.95 KB/s).
Source inspection at rust-v0.147.0 still shows the dedicated log DB using its own default TRACE filter. The current documented configuration has no setting that independently raises or disables this SQLite sink; RUST_LOG controls a different layer.
Updating alone is therefore not a discriminator here: the writer and latest stable release are both 0.147.0.

As a local stop-loss only, after stopping and independently verifying zero Codex handles on the DB/WAL/SHM, I installed one persistent BEFORE INSERT trigger on the dedicated logs table. It uses RAISE(IGNORE) only when upper(NEW.level) IN ('TRACE', 'DEBUG'); INFO/WARN/ERROR and all existing rows remain intact. No database was deleted, truncated, vacuumed, checkpointed, relocated, or read for content. Rollback is the exact named DROP TRIGGER operation.

An isolated truth-table/rollback test and live schema plus quick_check acceptance passed. After restart through the official managed daemon, a representative 30-second active sample wrote 217,088 bytes (7.24 KB/s), below the smaller report's 24.86 KB/s threshold and materially below both red rates. WAL metadata still changes because higher-severity diagnostics remain enabled.

This is not presented as an upstream fix or maintainer-endorsed workaround. It may disappear after a DB migration/recreation. A supported persisted log-level control (or an INFO-or-higher sink default), with migration behavior documented, would remove the need for this schema-level containment.

macOS: Persistent SQLite TRACE target=log churn remains after rust-v0.142.0 #29532

Description

Summary

Environment

Files observed

Read-only 60 second sample

Recent persisted targets

Expected behavior

Actual behavior

Privacy note

Activity

github-actions commented on Jun 23, 2026

zyfasos commented on Jun 23, 2026

vadxq commented on Jun 23, 2026

lock-down commented on Jun 23, 2026

nomideus commented on Jun 23, 2026

RonZhu commented on Jun 23, 2026

Environment

Before manual cleanup

After manual cleanup / compaction

Expected

Actual

mdbecker commented on Jun 23, 2026

nicksenround1 commented on Jun 24, 2026

Environment

Before cleanup

Repro after cleanup / restart

User-visible impact

zyfasos commented on Jun 24, 2026

Don-Quijote224 commented on Jun 24, 2026

almostlover-hao commented on Jun 24, 2026

Environment

Config tried

Observed behavior before mitigation

Why file size alone hides the issue

Local mitigation tested

Expected fix

Privacy note

20 remaining items

zyfasos commented on Jul 3, 2026

RonZhu commented on Jul 6, 2026

tangleo203 commented on Jul 10, 2026

Additional reproduction on ChatGPT-integrated Codex / 0.144.0-alpha.4

Environment

Evidence collected before installing any local workaround

Local confirmation/workaround

RonZhu commented on Jul 13, 2026

0xdevalias commented on Jul 13, 2026

Freyliu0516 commented on Jul 14, 2026

Proposed upstream change

Validation

User-side mitigation

Request

fujiwarakasei commented on Jul 15, 2026

Environment

Controlled test

Why I am posting

friday-james commented on Jul 16, 2026

Measurements: the writes are flush overhead, not content

Patch: flush by bytes/severity/60 s instead of every 2 s

A/B benchmark (each phase an isolated process)

papperrollinggery commented on Jul 16, 2026

Request for clarification: current shipping build still lacks the broader SQLite fix

RonZhu commented on Jul 20, 2026

TheJhyeFactor commented on Jul 20, 2026

RonZhu commented on Jul 23, 2026

bbingz commented on Jul 24, 2026

Analysis (community)

Observed (from issue + public residual signal)

Local residual measurement (2026-07-24, this reporter)

Signal-14 INTERIM evidence (from qwen_s14 mid-flight)

level counts (ro)

Root-cause hypothesis (to be confirmed on main)

High-level fix outline (not a PR)

Tests

Questions for maintainers

bbingz commented on Jul 24, 2026

Analysis (community)

Deterministic evidence (code + runtime)

Root-cause hypothesis

Root-cause hypothesis (to be confirmed on `main`)