Stop treating import calls as import system frames #329
Conversation
This file contains 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
Our existing flamegraph reporting code is subtly dependent upon the order in which allocations are passed to it. Add a test that illustrates this bug. This test is currently expected to fail, and will be made to pass in the next commit. Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
Up until now, our code has been treating the last location before the
import system is entered as an import system location as well, in order
to hide the `import` statement itself when the user clicks the "Hide
Import System Frames" checkbox.
There's a few problems with that approach, though. For one, there may be
other calls on the same line that lead to other, non-import system
allocations. For instance, the user may do something like:
import mmap; mmap.mmap(-1, 4096)
We don't want to hide that line on the flame graph, or users will lose
visibility into the `mmap.mmap` call when they try to hide imports.
A subtler problem with this approach is that we only handled the frame
with the `import` statement as an import system frame if the first child
we saw from it was an import system frame. Assuming it had a mix of some
descendants that are in the import system and some that aren't, our
behavior would change based on the order that the reporter received
those in (and the order that the reporter receives allocations in is
arbitrary and not guaranteed to be stable across runs).
Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
Once we've found an import system frame, we don't need to check any other frames in that call stack, since everything below an `import` call is part of that import. Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
Rather than calling `is_frame_from_import_system` every time we encounter a node, call it only when we first encounter a new node and add it to the tree. When a call stack reuses a previously seen node, also reuse its `is_import_system` flag. Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
If a node has previously been added to the tree, we know that it isn't CPython internal (or else we would have skipped adding it). Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
This is gonna sound stupid, but: `py-spy` tells me that around 6% of our flamegraph generation runtime in Python 3.10 is saved if we don't use star unpacking on this hot path, and instead depend upon `frame` having exactly 3 fields. Signed-off-by: Matt Wozniski <mwozniski@bloomberg.net>
godlygeek
force-pushed
the
fix_import_system_hiding
branch
from
725c40d
to
de70269
Compare
pablogsal
approved these changes
Mar 6, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Up until now, our code has been treating the last location before the
import system is entered as an import system location as well, in order
to hide the
importstatement itself when the user clicks the "HideImport System Frames" checkbox.
There's a few problems with that approach, though. For one, there may be
other calls on the same line that lead to other, non-import system
allocations. For instance, the user may do something like:
We don't want to hide that line on the flame graph, or users will lose
visibility into the
mmap.mmapcall when they try to hide imports.A subtler problem with this approach is that we only handled the frame
with the
importstatement as an import system frame if the first childwe saw from it was an import system frame. Assuming it had a mix of some
descendants that are in the import system and some that aren't, our
behavior would change based on the order that the reporter received
those in (and the order that the reporter receives allocations in is
arbitrary and not guaranteed to be stable across runs).
Fix this by never hiding the
importstatement itself.Also, add some micro optimizations here. These optimizations speed up
flamegraph generation by about 15% (for the nbody example, native mode).