Fix: replace verbose log tiers with severity levels - #1475
Merged
ChaoZheng109 merged 1 commit intoJul 29, 2026
Conversation
|
Important Review skippedToo many files! This PR contains 161 files, which is 61 over the limit of 100. To get a review, narrow the scope: Upgrade to a paid plan to raise the limit. This review couldn't start because sufficient usage credits or metered capacity aren't available. Add credits or update usage-based reviews in the billing tab, then retry. ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (161)
You can disable this status message by setting the Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
zmnobug
force-pushed
the
refactor/issue-1429-log-levels
branch
4 times, most recently
from
July 27, 2026 03:36
873df6e to
2d4b378
Compare
6 tasks
ChaoWao
added a commit
that referenced
this pull request
Jul 27, 2026
docs/ had 35 files flat at its root with no index. The top-level README listed 10 of them, so the other 25 were discoverable only by grep, and four had no inbound link from anywhere in the repo. Only investigations/ carried a README; dfx/ (12 files, 5227 lines), troubleshooting/ and hardware/ had none. - Add docs/README.md indexing every root document and subdirectory, grouped by reader intent, and stating that a new doc registers there in the same commit. This de-orphans install.md, simt-launch.md, profiling-config-naming.md and sim_multi_device_isolation.md. - Add docs/capability-survey.md: a status snapshot of what is shipped, gated, or design-only across the level model (L2 chip up to L4 host-to-host), the CANN launch surface for AICore and AICPU, and the communication engines. Every claim carries a file:line so it can be re-checked, and the page records the snapshot commit because status claims rot faster than architecture. - Move profiling-framework.md, profiling-name-map.md and profiling-config-naming.md into docs/dfx/, which already held 12 profiling documents and linked up to these three nine times via ../. Add docs/dfx/README.md as that directory's index. logging.md stays at the docs/ root: PR #1475 modifies it, so renaming it here would hand that author a rename-vs-modify conflict. - Fix seven broken links in docs/dfx/pmu-profiling.md, a file already touched by this change: six had the wrong relative depth (../src -> ../../src) and profiling_copy.h moved to src/common/ when #944 unified the collector framework.
8 tasks
zmnobug
force-pushed
the
refactor/issue-1429-log-levels
branch
4 times, most recently
from
July 28, 2026 11:08
accb802 to
45d062e
Compare
Collaborator
|
Stale reference to a removed symbol — # line 25
from simpler import _log
# line 69
logging.getLogger("simpler").setLevel(_log.V9)This PR deletes The file isn't part of this PR's diff, so a line-anchored inline comment can't be attached here — flagging it at the conversation level instead. It's also why CI stays green: the filename isn't Suggested fix: |
zmnobug
force-pushed
the
refactor/issue-1429-log-levels
branch
from
July 29, 2026 06:23
45d062e to
812abe1
Compare
Contributor
Author
|
@ChaoZheng109 Addressed the stale benchmark reference: |
- Collapse V0-V9 call sites to INFO and reserve TIMING for STRACE - Add DEBUG, TIMING, WARN, and ERROR severity APIs - Default to TIMING and align Python, host, sim, and onboard thresholds - Gate onboard TIMING independently from CANN's coarser WARN level - Forward the threshold into RTLD_LOCAL AICPU simulation libraries - Add CANN configuration and host/device propagation regressions - Update affected call sites and documentation Fixes hw-native-sys#1429 Co-authored-by: ChaoZheng109 <zhengchao47@huawei.com>
zmnobug
force-pushed
the
refactor/issue-1429-log-levels
branch
from
July 29, 2026 07:08
812abe1 to
b37da08
Compare
4 tasks
Collaborator
Verification: functional (per-level) + performance (#1409 benefit)Tested on real a2a3 silicon (
1. Functional — every
|
| level | STRACE | DEBUG | INFO | TIMING |
|---|---|---|---|---|
| debug | ✓ | ✓ | ✓ | ✓ |
| info | ✓ | ✗ | ✓ | ✓ |
| timing (default) | ✓ | ✗ | ✗ | ✓ |
| warn | ✗ | ✗ | ✗ | ✗ |
| error / null | (silent) |
Default timing keeps host STRACE while suppressing INFO — as intended.
a2a3 onboard, all 6 levels, 6/6 passed. The debug/ (applog) device stream — the one dlog_setlevel controls — is gated exactly right: info → 266 INFO/round, timing/warn → 0, null → empty file.
One nuance worth recording: a fixed ~20 INFO lines remain in the
run/device partition (device ccpu system process run log, from the residentaicpu_schedulerdaemon:ae_so_manager GetApi, TSD events). These are invariant across--log-level(timing→null) andASCEND_DEVICE_LOG_LEVEL=3— verified empirically. They are not reachable bydlog_setlevel(which governs applog/debuglog only), so this is not a gap in this PR; simpler mutes everything the applog level mechanism can mute.
2. Performance — the #1409 device-INFO tax is still eliminated
batch_paged_attention Case1, 100 rounds, same die, A-B-A (timing → info → timing) to bound drift:
| run | level | Avg Device (us) | Avg Effective (us) |
|---|---|---|---|
| A | timing | 4247.7 | 4223.0 |
| B | info | 4810.0 | 4424.4 |
| C | timing | 4314.5 | 4290.0 |
timing baseline (A,C mean): Device 4281.1, Effective 4256.5.
| metric | timing → info penalty |
|---|---|
| Device (device_wall) | −528.9 us / −11.0% |
| Effective (orch∪sched) | −167.9 us / −3.8% |
Drift between the two timing runs is only 66.8 us (1.6%) — far below the info penalty, so the signal is robust. Effective −3.8% closely matches this PR's reported −4.67%; Device shows an even larger gap here (my absolute info-Device is higher, likely shared-box contention during the info run). Mechanism confirmed: info opens ~266 device INFO/round (the 6299-over-100-rounds tax) that land inside the device wall; default timing drives it to 0.
Conclusion: all --log-level values are functional, host/device stay aligned, and the #1409 optimization is intact (default timing avoids the device-INFO tax; ~−3.8% Effective, in line with the PR).
ChaoZheng109
approved these changes
Jul 29, 2026
ChaoZheng109
added a commit
to ChaoZheng109/simpler
that referenced
this pull request
Jul 29, 2026
The scalar_data orchestration migrated LOG_INFO_V0 to LOG_INFO when the verbose V0..V9 tiers were replaced by severity levels (hw-native-sys#1475), but its README still named the removed macro. Align the doc with the code. Co-authored-by: zm <zhaomin88@huawei.com> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ChaoZheng109
added a commit
to ChaoZheng109/simpler
that referenced
this pull request
Jul 29, 2026
The scalar_data orchestration migrated LOG_INFO_V0 to LOG_INFO when the verbose V0..V9 tiers were replaced by severity levels (hw-native-sys#1475), but its README still named the removed macro. Align the doc with the code. Co-authored-by: zm <zhaomin88@huawei.com>
ChaoZheng109
added a commit
that referenced
this pull request
Jul 29, 2026
The scalar_data orchestration migrated LOG_INFO_V0 to LOG_INFO when the verbose V0..V9 tiers were replaced by severity levels (#1475), but its README still named the removed macro. Align the doc with the code. Co-authored-by: zm <zhaomin88@huawei.com>
4 tasks
ChaoWao
added a commit
to ChaoWao/simpler-fork
that referenced
this pull request
Jul 29, 2026
`logging.md` opens by pointing at `testing.md § Log levels` for the user-facing model and claiming implementation for itself; `testing.md § Log levels` points back for implementation. The split is declared in both directions, and neither file held it — testing.md carried the `simpler_log_init` / `HostLogger` / `InitArgs` chain, the `cann_level()` mapping, and the fork-time threshold propagation, all of which are mechanism, and logging.md restated the snapshot semantics. testing.md now answers only what a test author needs: what the levels are, which macro lands where, how to set the threshold, and the two consequences — the threshold is snapshotted once at `Worker.init()`, and onboard AICPU severity comes from CANN, where a pre-set `ASCEND_GLOBAL_LOG_LEVEL` wins. logging.md gains what moved: the CANN mapping as a table under Configuration flow, plus the forked-subprocess rule, which was documented nowhere else — the parent snapshots before `fork()` and passes the value explicitly to `_chip_process_loop` rather than letting the child re-read a logger it never configured. Both files now also state the deadline for configuring CANN severity by hand: it must happen before `Worker.init()`. `init_log_switch()` reads CANN once at the top of the AICPU init kernel and latches the result into `g_is_log_enable_*`; every later `LOG_*` tests those flags and never asks CANN again, and that kernel runs inside `Worker.init()`. A `dlog_setlevel` after that call silently does nothing to device-side logging. Not a size reduction: testing.md loses 6 lines and logging.md gains 37. The point is that one topic now has one owner per aspect, so a change like hw-native-sys#1475 has one place to land instead of two. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
ChaoWao
added a commit
that referenced
this pull request
Jul 30, 2026
`logging.md` opens by pointing at `testing.md § Log levels` for the user-facing model and claiming implementation for itself; `testing.md § Log levels` points back for implementation. The split is declared in both directions, and neither file held it — testing.md carried the `simpler_log_init` / `HostLogger` / `InitArgs` chain, the `cann_level()` mapping, and the fork-time threshold propagation, all of which are mechanism, and logging.md restated the snapshot semantics. testing.md now answers only what a test author needs: what the levels are, which macro lands where, how to set the threshold, and the two consequences — the threshold is snapshotted once at `Worker.init()`, and onboard AICPU severity comes from CANN, where a pre-set `ASCEND_GLOBAL_LOG_LEVEL` wins. logging.md gains what moved: the CANN mapping as a table under Configuration flow, plus the forked-subprocess rule, which was documented nowhere else — the parent snapshots before `fork()` and passes the value explicitly to `_chip_process_loop` rather than letting the child re-read a logger it never configured. Both files now also state the deadline for configuring CANN severity by hand: it must happen before `Worker.init()`. `init_log_switch()` reads CANN once at the top of the AICPU init kernel and latches the result into `g_is_log_enable_*`; every later `LOG_*` tests those flags and never asks CANN again, and that kernel runs inside `Worker.init()`. A `dlog_setlevel` after that call silently does nothing to device-side logging. Not a size reduction: testing.md loses 6 lines and logging.md gains 37. The point is that one topic now has one owner per aspect, so a change like #1475 has one place to land instead of two.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
DEBUG,INFO,TIMING,WARN, andERROR, defaulting toTIMINGTIMING, migrate affected call sites and documentation, and retain the log-footprint reduction from Strip all LOG_* call sites — logging re-tiering baseline #1447Testing
Performance
On a2a3
batch_paged_attentionCase1 over 100 rounds:Fixes #1429