profilers/python_ebpf.py: specifying newly TMPDIR on each instance of PyPerf #921
Conversation
gprofiler/profilers/python_ebpf.py
Outdated
| self._pyperf_staticx_tmpdir: Optional[Path] = None | ||
| if os.environ.get("TMPDIR", None) is not None: | ||
| # We want to create a new level of hirerachy in our current staticx tempdir. | ||
| self._pyperf_staticx_tmpdir = Path(os.environ["TMPDIR"]) / random_prefix() |
There was a problem hiding this comment.
| self._pyperf_staticx_tmpdir = Path(os.environ["TMPDIR"]) / random_prefix() | |
| self._pyperf_staticx_tmpdir = Path(os.environ["TMPDIR"]) / "pyperf_" + random_prefix() |
|
@d3dave are you able to complete the review this week? |
CI is red |
|
Merged the CI fix from |
| # ensure `TMPDIR` env is propagated to the child processes (used by staticx) | ||
| if "TMPDIR" not in env and "TMPDIR" in os.environ: | ||
| if tmpdir is not None: | ||
| tmpdir.mkdir(exist_ok=True) | ||
| env["TMPDIR"] = tmpdir.as_posix() | ||
| elif "TMPDIR" not in env and "TMPDIR" in os.environ: | ||
| # ensure `TMPDIR` env is propagated to the child processes (used by staticx) |
There was a problem hiding this comment.
I’m not sure this is the right approach
maybe should we use tmpdir always relative to the current env[“TMPDIR”] (e.g. f”{env[“TMPDIR”]}/{tmpdir}”)
There was a problem hiding this comment.
Is there a practical difference? Doesn't the current flow always create the tmpdir passed to this function under $TMPDIR?
There was a problem hiding this comment.
@d3dave current approach:
if tmpdir is set - mkdir it (w/o parents). propagate it to the child.
suggested approach:
if tmpdir is set - mkdir "$TMPDIR/tmpdir“. propagate it to the child.
There was a problem hiding this comment.
Ok, but the tmpdir arg, whenever it is passed it is always Path(os.environ["TMPDIR"]) / ("pyperf_" + random_prefix()) which is under $TMPDIR. So there is no difference because with both approaches the tmpdir is created under $TMPDIR.
There was a problem hiding this comment.
there’s no practical differences… I raised this to discuss what’s the better approach
Description
cleanup each instance of PyPerf staticx directories on termination.
Related Issue
We experienced where there were a-lot of staticx tempdirs leftovers.
Motivation and Context
How Has This Been Tested?
Screenshots
Checklist: