use valgrind --leak-check=yes to detect memory leak - Mailing list pgsql-hackers

From Alex
Subject use valgrind --leak-check=yes to detect memory leak
Date
Msg-id CAKU4AWoDby5gnp6nd1mMtsT+0vmHjjDyhuXjgGZTDcKsgxr-3g@mail.gmail.com
Whole thread Raw
Responses Re: use valgrind --leak-check=yes to detect memory leak
List pgsql-hackers
I want to use valgrind to detect memory leak issue.   Then I run into 2 problems I want to confirm them here. 

Q1:  can I use 'leak-check=yes' to detect memory leak issue?
1.  In https://wiki.postgresql.org/wiki/Valgrind,   the wiki indicates to use '--leak-check=no'
2.  in https://github.com/afiskon/pgscripts/blob/master/valgrind.sh#L55,  it use 'leak-check=no' as well with the wold "No point to check for memory leaks, Valgrind doesn't understand MemoryContexts and stuff"
3.  Per my understanding,  if we build pg with USE_VALGRIND, then we can use 'leak-check=yes' to detect memory leak issue,  the idea here is  
     a).  valgrind can't understand MemoryContext, 
     b).  with USE_VALGRIND,  pg use ` VALGRIND_MEMPOOL_ALLOC` and ` VALGRIND_MEMPOOL_FREE`  whenever we we allocate and free memory in MemoryContext
     c).  finally, valgrind check the valgrind_mempool to know which memory is leak. 
So the answer should be "yes, we can use "leak-check=yes" to detect memory leak issue?


Q2:   do we check memory leak for some new commits or we can ignore them based on we use memory context carefully?  If we want to check memory leak for some new commits,  how to handle the existing memory leak case?

with `valgrind --leak-check=yes --suppressions=src/tools/valgrind.supp ...` and run `make installcheck` on an unmodified version (commit d06fe6ce2c79420fd19ac89ace81b66579f08493) ,  I run into 711 memory leaks with `match-leak-kinds: definite`.   if we want to check memory leak for new commits, this should be a kind of troubles.  how do you do with this everyday?

Thanks

Attachment

pgsql-hackers by date:

Previous
From: David Rowley
Date:
Subject: Re: Custom table AMs need to include heapam.h because of BulkInsertState
Next
From: Andres Freund
Date:
Subject: Re: POC: Cleaning up orphaned files using undo logs