Lock internal and principal keys in RAM #279
Merged
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
dAdAbird
force-pushed
the
mlock
branch
2 times, most recently
from
9b3dfca to
da4718d
Compare
dAdAbird
changed the title
This commit makes internal keys cache data being locked in RAM to prevent this data from being paged to the swap. As memory locking is performed in units of whole pages, this commit also redesigns the cache so all records are compactly placed in pages. Before it was a linked list with nodes in random places. So locking each key would mean potential "wasting" of the locked page as its number is limited (although on modern systems it is a fairly big number) and we can't be sure that the next record would be on the same page. Also, having records sequentially placed makes iterations through them CPU cache friendly in contrast to random memory pointers of linked lists.
We just lock the hash table entry. Although it will lock the whole mem page we ok with that potential waste. The number of principal keys should be less than internal and allocated in the shared memory once for all backends (unlike internal keys in the TopMemoryContext of every backend). Plus hash table (via dsa) allocates in 4kb pages and hopefully, principal keys end up compactly placed.
dutow
approved these changes
Sep 17, 2024
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Outdated suggestions cannot be applied.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.
This PR makes internal and principal keys cache data being locked in RAM to prevent this data from being paged to the swap.
For the internal keys, as memory locking is performed in units of whole pages, this commit also redesigns the cache so all records are compactly placed in pages. Before it was a linked list with nodes in random places. So locking each key would mean potential "wasting" of the locked page as its number is limited (although on modern systems it is a fairly big number) and we can't be sure that the next record would be on the same page. Also, having records sequentially placed makes iterations through them CPU cache friendly in contrast to random memory pointers of linked lists.
For the principal keys, we just lock the hash table entry. Although it will lock the whole mem page we ok with that potential waste. The number of principal keys should be less than internal and allocated in the shared memory once for all backends (unlike internal keys in the TopMemoryContext of every backend). Plus hash table (via dsa) allocates in 4kb pages and hopefully, principal keys end up compactly placed.
For https://perconadev.atlassian.net/browse/PG-823