Repository navigation
Memory leak in dict with shared keys #75498
Description
| BPO | 31317 |
|---|---|
| Nosy | @rhettinger, @pitrou, @vstinner, @methane, @serhiy-storchaka, @zhangyangyu |
| Superseder | |
| Files |
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
assignee = None
closed_at = <Date 2017-08-31.22:23:00.056>
created_at = <Date 2017-08-31.20:18:11.860>
labels = ['3.7', 'performance']
title = 'Memory leak in dict with shared keys'
updated_at = <Date 2017-08-31.22:23:00.054>
user = 'https://github.com/vstinner'bugs.python.org fields:
activity = <Date 2017-08-31.22:23:00.054>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2017-08-31.22:23:00.056>
closer = 'vstinner'
components = []
creation = <Date 2017-08-31.20:18:11.860>
creator = 'vstinner'
dependencies = []
files = ['47111']
hgrepos = []
issue_num = 31317
keywords = []
message_count = 8.0
messages = ['301058', '301060', '301062', '301063', '301064', '301065', '301067', '301068']
nosy_count = 6.0
nosy_names = ['rhettinger', 'pitrou', 'vstinner', 'methane', 'serhiy.storchaka', 'xiang.zhang']
pr_nums = []
priority = 'normal'
resolution = 'duplicate'
stage = 'resolved'
status = 'closed'
superseder = '31217'
type = 'resource usage'
url = 'https://bugs.python.org/issue31317'
versions = ['Python 3.7']
Activity
vstinner commentedon Aug 31, 2017
Copy attached test_leak.py to Lib/test/ and run:
haypo@selma$ ./python -m test -R 3:30 test_leak
Run tests in parallel using 1 child processes
0:00:00 load avg: 0.67 [1/1/1] test_leak failed
beginning 33 repetitions
123456789012345678901234567890123
.................................
test_leak leaked [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] memory blocks, sum=30
1 test failed:
test_leak
Total duration: 682 ms
Tests result: FAILURE
It seems like Thread2 constructor creates a dictionary on "self.hashvalue = (...)". It's a dictionary sharing keys, but it seems like the dict table memory is not released whne the bug occurs.
Sadly, I'm only able to reproduce the bug on a 32-bit build. Since the bug depends on exact hash values, maybe the bug exists in 64-bit builds, but is less likely?
I'm not 100% sure if the bug is related to dict or set.
The bug comes from bpo-31217.
vstinner commentedon Aug 31, 2017
If you cannot reproduce the bug on 32-bit build, try to set "FIXED_HASH = 0" in test_leak.py, run "./python -m test -R 3:10 test_leak" multiple times until it fails, then copy the last value of the printed list into hash_values in test_leak.py.
vstinner commentedon Aug 31, 2017
I spent a lot of time in gdb, but I'm not sure anymore that the issue is related to dict shared keys.
pitrou commentedon Aug 31, 2017
It's a really weird bug you found. It depends on the exact values passed to -R:
pitrou commentedon Aug 31, 2017
Actually, it's not the bug that is weird, but the heuristics in -R seem completely weird now. If I print the alloc and rc deltas:
pitrou commentedon Aug 31, 2017
s/completely weird/broken/
pitrou commentedon Aug 31, 2017
To come back to this bug, it still fails if add a __slots__, so I don't think it's related to shared key dictionaries.
vstinner commentedon Aug 31, 2017
Antoine: "Actually, it's not the bug that is weird, but the heuristics in -R seem completely weird now. If I print the alloc and rc deltas:"
I changed it recently to reduce false alarms: see bpo-30776.
Antoine: "To come back to this bug, it still fails if add a __slots__, so I don't think it's related to shared key dictionaries."
Oh ok, in that case, I will move back to bpo-31217. I close this issue as a duplicate.