ChainDB: batch garbage collections #1932
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
Previously, we scheduled a garbage collection for each block for 10 seconds in the future. This meant that our scheduled GCs queue was blocks/s * 10 long. When tracing the queue length on my machine, it hovered between 5000 and 6000 entries. Moreover, a VolatileDB garbage collection is triggered at blocks/s, which should result in a lot of contention for the VolatileDB state. Even worse is that a 10 second delay is too short to reliably ensure the block will have been flushed to disk (in the ImmutableDB) before it is garbage collected. However, increasing this delay would make the queue significantly longer. To fix these issues, we introduce a GC interval (in seconds). We batch all GCs in the same interval together. This means that the queue length is now at most ⌈delay / interval⌉ + 1, e.g., 60s / 10s = 7, which is much shorter than 5000-6000. Moreover, there will be at most one GC every `interval` seconds, e.g., 10s. The cost of switching to a longer GC delay is that the in-memory index of the VolatileDB will be larger, making operations on it, such as lookups, more expensive (most operations are `O(n*log(n))`). See the docstring of 'defaultSpecificArgs' for what the new default values of `gcDelay` and `gcInterval` mean in practice.
edsko
approved these changes
Apr 10, 2020
|
bors merge |
coot
pushed a commit
that referenced
this pull request
May 16, 2022
1932: ChainDB: batch garbage collections r=mrBliss a=mrBliss Previously, we scheduled a garbage collection for each block for 10 seconds in the future. This meant that our scheduled GCs queue was blocks/s * 10 long. When tracing the queue length on my machine, it hovered between 5000 and 6000 entries. Moreover, a VolatileDB garbage collection is triggered at blocks/s, which should result in a lot of contention for the VolatileDB state. Even worse is that a 10 second delay is too short to reliably ensure the block will have been flushed to disk (in the ImmutableDB) before it is garbage collected. However, increasing this delay would make the queue significantly longer. To fix these issues, we introduce a GC interval (in seconds). We batch all GCs in the same interval together. This means that the queue length is now at most ⌈delay / interval⌉ + 1, e.g., 60s / 10s = 7, which is much shorter than 5000-6000. Moreover, there will be at most one GC every `interval` seconds, e.g., 10s. The cost of switching to a longer GC delay is that the in-memory index of the VolatileDB will be larger, making operations on it, such as lookups, more expensive (most operations are `O(n*log(n))`). See the docstring of 'defaultSpecificArgs' for what the new default values of `gcDelay` and `gcInterval` mean in practice. Co-authored-by: Thomas Winant <[email protected]>
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.
Previously, we scheduled a garbage collection for each block for 10 seconds in
the future. This meant that our scheduled GCs queue was blocks/s * 10 long.
When tracing the queue length on my machine, it hovered between 5000 and 6000
entries. Moreover, a VolatileDB garbage collection is triggered at blocks/s,
which should result in a lot of contention for the VolatileDB state.
Even worse is that a 10 second delay is too short to reliably ensure the block
will have been flushed to disk (in the ImmutableDB) before it is garbage
collected. However, increasing this delay would make the queue significantly
longer.
To fix these issues, we introduce a GC interval (in seconds). We batch all GCs
in the same interval together. This means that the queue length is now at most
⌈delay / interval⌉ + 1, e.g., 60s / 10s = 7, which is much shorter than
5000-6000. Moreover, there will be at most one GC every
intervalseconds,e.g., 10s.
The cost of switching to a longer GC delay is that the in-memory index of the
VolatileDB will be larger, making operations on it, such as lookups, more
expensive (most operations are
O(n*log(n))). See the docstring of'defaultSpecificArgs' for what the new default values of
gcDelayandgcIntervalmean in practice.