[PyROOT] Fix a GIL deadlock in THx.Fit#21129
[PyROOT] Fix a GIL deadlock in THx.Fit#21129hageboeck wants to merge 4 commits intoroot-project:masterfrom
Conversation
TH1.Fit might call into Python functions, which would require the GIL to be held. If TH1.Fit holds it, though, the process deadlocks. Here, the GIL is released before TH1.Fit, which runs fully in C++ anyway. Fix root-project#21080.
This saves startup times and collects TH1 tests in one place.
The timeout of the test is reduced because the deadlock would otherwise show only after 1500s.
hageboeck
requested review from
bellenot,
guitargeek and
vepadulano
as code owners
|
Nice, and compute intense functions like these is exactly when the GIL should be released to enable Python multithreading. Did you check the Pythonization also works for TH1-derived classes? |
We checked in the interpreter: >>> import ROOT
>>> ROOT.TH2.Fit.__release_gil__
True |
Test Results 22 files 22 suites 3d 14h 15m 35s ⏱️ For more details on these failures, see this check. Results for commit 0319cda. ♻️ This comment has been updated with latest results. |
|
I've taken a look at the MacOS failures. There seems to be some other kind of interference on that platform for two different types of functions that release the GIL, namely the distributed RDataFrame execution and now TH1.Fit. Needs investigation |
vepadulano
left a comment
vepadulano
left a comment
There was a problem hiding this comment.
Thank you @siliataider for investigating and fixing the failures on MacOS!
The changes LGTM, maybe consider the ruff format suggestions.
4 participants
TH1.Fit might call into Python functions, which would require the GIL to
be held. If TH1.Fit holds it, though, the process deadlocks.
Here, the GIL is released before TH1.Fit, which runs fully in C++
anyway.
A test was added with a 30s timeout. It completes in about 1s, unless the
GIL deadlock is maintained.
Fix #21080.