refactor(internal): use a single universal finder for module watchdogs - #19790
Open
P403n1x87 wants to merge 1 commit into
Open
refactor(internal): use a single universal finder for module watchdogs#19790P403n1x87 wants to merge 1 commit into
P403n1x87 wants to merge 1 commit into
Conversation
BaseModuleWatchdog subclasses (ModuleWatchdog, DebuggerModuleWatchdog,
ModuleCodeCollector, SymbolDatabaseUploader, and the errortracking
watchdogs) each inserted themselves into sys.meta_path and, on every
find_spec/find_module call, scanned the rest of sys.meta_path to find
the real underlying finder. With N watchdogs installed simultaneously
(realistic in production: debugger + coverage + symbol DB + error
tracking can all be enabled on one service), this produced O(N^2)
find_spec invocations per import.
A standalone benchmark against the unrefactored code confirmed the
call count follows the exact triangular number N(N+1)/2 (e.g. 21
calls for N=6, where 1 would do), and wall-clock import time grew
correspondingly (~1.95x baseline at N=6, unbounded as feature count
grows). Module imports sit on every process-startup hot path, so this
was worth fixing.
This introduces _UniversalModuleWatchdog: the single object ever
inserted into sys.meta_path for watchdog purposes, mirroring the
existing _UniversalWrappingContext/WrappingContext pattern in
ddtrace/internal/wrapping/context.py. Concrete watchdog subclasses
keep their own singleton instance and state (still read directly as
production API, e.g. ModuleCodeCollector._instance), but register as
participants with the one real finder instead of installing their
own. The real find_spec/find_module lookup now happens once per
import regardless of N; only the O(N) per-participant callback/
transformer registration remains.
Re-running the adapted benchmark against the refactored code confirms
find_spec calls are now flat at 1 per import for N in {1..30}, versus
the prior triangular growth.
As part of unifying hook resolution through the single participant
list, _pre_exec_module_hooks/_import_exception_hooks resolution now
uses first-registered-wins (iterating _participants in registration
order) instead of the previous most-recently-installed-wins (an
accident of meta_path insertion order). after_import/transform
callbacks are unaffected: they already fired in registration order.
Also wraps each after_import callback invocation in try/except so one
misbehaving hook can't prevent the rest from running, folding in a
fix that was previously written but unmerged on
chore/handle-after-import-exceptions.
Circular import analysis
|
Dependency direction analysis
|
Contributor
|
P403n1x87
added
the
changelog/no-changelog
Codeowners resolved asResolved from the full PR diff against |
BenchmarksBenchmark execution time: 2026-08-20 14:00:08 Comparing candidate commit 19ac065 in PR branch Found 0 performance improvements and 11 performance regressions! Performance is the same for 605 metrics, 11 unstable metrics.
|
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
1 participant
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.
Description
BaseModuleWatchdog subclasses (ModuleWatchdog, DebuggerModuleWatchdog, ModuleCodeCollector, SymbolDatabaseUploader, and the errortracking watchdogs) each inserted themselves into sys.meta_path and, on every find_spec/find_module call, scanned the rest of sys.meta_path to find the real underlying finder. With N watchdogs installed simultaneously (realistic in production: debugger + coverage + symbol DB + error tracking can all be enabled on one service), this produced O(N^2) find_spec invocations per import.
A standalone benchmark against the unrefactored code confirmed the call count follows the exact triangular number N(N+1)/2 (e.g. 21 calls for N=6, where 1 would do), and wall-clock import time grew correspondingly (~1.95x baseline at N=6, unbounded as feature count grows). Module imports sit on every process-startup hot path, so this was worth fixing.
This introduces _UniversalModuleWatchdog: the single object ever inserted into sys.meta_path for watchdog purposes, mirroring the existing _UniversalWrappingContext/WrappingContext pattern in ddtrace/internal/wrapping/context.py. Concrete watchdog subclasses keep their own singleton instance and state (still read directly as production API, e.g. ModuleCodeCollector._instance), but register as participants with the one real finder instead of installing their own. The real find_spec/find_module lookup now happens once per import regardless of N; only the O(N) per-participant callback/ transformer registration remains.
Re-running the adapted benchmark against the refactored code confirms find_spec calls are now flat at 1 per import for N in {1..30}, versus the prior triangular growth.
As part of unifying hook resolution through the single participant list, _pre_exec_module_hooks/_import_exception_hooks resolution now uses first-registered-wins (iterating _participants in registration order) instead of the previous most-recently-installed-wins (an accident of meta_path insertion order). after_import/transform callbacks are unaffected: they already fired in registration order.