Handle merge of already serialized value-keys #95
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
willscott
approved these changes
Oct 4, 2022
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## main #95 +/- ##
==========================================
- Coverage 63.57% 63.47% -0.11%
==========================================
Files 15 15
Lines 2740 2779 +39
==========================================
+ Hits 1742 1764 +22
- Misses 742 753 +11
- Partials 256 262 +6
|
Pebble documentation suggests that `Merger` is used by `.Merge` calls on a given key value. However, Pebble uses merger on iteration over values too. Internally, `Get` operations use iterators for example. This means values that are already serialized may be passed to the merger implementation during in-memory traversal. The previous implementation of value-keys merger did not take this into account and as a result it meant that `Get` failed to find value-keys associated to a multihash since they ended up being double wrapped by the merger value-keys marshaller. The changes here fixes this issue by checking explicitly that a give value-key has the expected key prefix before proceeding with the remaining merge logic. As a precaution, the merger is also modified to fall back on the default Pebble merger when the key is not of kind multihash-key. Further, the benchmarks are modified to explicitly assert that the inserted values are indeed found for each entry. The `Get` implementation in Pebble store is also modified to copy the content returned by the backing store to avoid slice invalidation and runtime invalid reference errors captured in #80. Unrelated to the lookup issue, the changes here also fix typos in benchmarks where workload does not match the benchmark name. Further, the common tests are updated to do a flush as suggested in #93. Fixes #94, #80.
masih
force-pushed
the
masih/pebble_merger_fix
branch
from
2083f1b to
d5b60db
Compare
|
For reference here is a local run of $ go test ./bench -bench='(Pebble|Storethehash)(Put|Get)_W(0|1)' -count=3 -timeout=300m
cpu: Intel(R) Core(TM) i7 CPU @ 2.30GHz
PASS
ok github.com/filecoin-project/go-indexer-core/bench 1416.652s
name time/op
Store_PebblePut_W0-8 1.78s ±12%
Store_PebbleGet_W0-8 9.03s ± 1%
Store_PebblePut_W1-8 21.4s ± 6%
Store_PebbleGet_W1-8 135s ± 3%
Store_StorethehashPut_W0-8 3.38s ± 4%
Store_StorethehashGet_W0-8 9.82s ± 4%
Store_StorethehashPut_W1-8 55.9s ± 1%
Store_StorethehashGet_W1-8 141s ± 9%
name speed
Store_PebblePut_W0-8 19.3MB/s ±11%
Store_PebbleGet_W0-8 3.77MB/s ± 1%
Store_PebblePut_W1-8 15.9MB/s ± 6%
Store_PebbleGet_W1-8 2.53MB/s ± 3%
Store_StorethehashPut_W0-8 10.1MB/s ± 4%
Store_StorethehashGet_W0-8 3.47MB/s ± 4%
Store_StorethehashPut_W1-8 6.10MB/s ± 1%
Store_StorethehashGet_W1-8 2.43MB/s ± 9%
name size
Store_PebblePut_W0-8 48.4M ± 0%
Store_PebbleGet_W0-8 45.5M ± 0%
Store_PebblePut_W1-8 501M ± 0%
Store_PebbleGet_W1-8 487M ± 0%
Store_StorethehashPut_W0-8 87.4M ± 0%
Store_StorethehashGet_W0-8 87.4M ± 0%
Store_StorethehashPut_W1-8 912M ± 0%
Store_StorethehashGet_W1-8 912M ± 0%
name alloc/op
Store_PebblePut_W0-8 155MB ± 1%
Store_PebbleGet_W0-8 19.6GB ± 0%
Store_PebblePut_W1-8 4.51GB ± 2%
Store_PebbleGet_W1-8 198GB ± 0%
Store_StorethehashPut_W0-8 967MB ± 0%
Store_StorethehashGet_W0-8 1.61GB ± 5%
Store_StorethehashPut_W1-8 11.1GB ± 1%
Store_StorethehashGet_W1-8 17.8GB ± 0%
name allocs/op
Store_PebblePut_W0-8 2.55M ± 2%
Store_PebbleGet_W0-8 29.7M ± 0%
Store_PebblePut_W1-8 97.9M ± 2%
Store_PebbleGet_W1-8 351M ± 1%
Store_StorethehashPut_W0-8 11.9M ± 1%
Store_StorethehashGet_W0-8 28.1M ± 5%
Store_StorethehashPut_W1-8 142M ± 0%
Store_StorethehashGet_W1-8 309M ± 0% |
gammazero
approved these changes
Oct 4, 2022
Make a copy of given byte slices to assure they remain valid and are unaffected by slice invalidation.
masih
force-pushed
the
masih/pebble_merger_fix
branch
from
521d8bc to
2e1c674
Compare
masih
added a commit
that referenced
this pull request
Oct 6, 2022
The previous behaviour of value-key merger in certain scenarios may have been resulted in value-keys being marshalled repeatedly. This was the underlying cause of the issue captured in #94 and corrected in #95. The work here adds to the corrections by gracefully fixing such records by unmarshalling them recursively until we reach values with the expected key prefix, and de-duplicating them as necessary. The fix is done opportunistically, in that whenever such values are detected during reads and writes they are fixed. Additional tests are added to simulate nested marshalling of value-keys and assert that the resulting merge is as expected for both merge-older and merge-newer cases.
masih
added a commit
that referenced
this pull request
Oct 6, 2022
The previous behaviour of value-key merger in certain scenarios may have been resulted in value-keys being marshalled repeatedly. This was the underlying cause of the issue captured in #94 and corrected in #95. The work here adds to the corrections by gracefully fixing such records by unmarshalling them recursively until we reach values with the expected key prefix, and de-duplicating them as necessary. The fix is done opportunistically, in that whenever such values are detected during reads and writes they are fixed. Additional tests are added to simulate nested marshalling of value-keys and assert that the resulting merge is as expected for both merge-older and merge-newer cases.
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.
Pebble documentation suggests that
Mergeris used by.Mergecalls on a given key value. However, Pebble uses merger on iteration over values too. Internally,Getoperations use iterators for example. This means values that are already serialized may be passed to the merger implementation during in-memory traversal.The previous implementation of value-keys merger did not take this into account and as a result it meant that
Getfailed to find value-keys associated to a multihash since they ended up being double wrapped by the merger value-keys marshaller.The changes here fixes this issue by checking explicitly that a give value-key has the expected key prefix before proceeding with the remaining merge logic. As a precaution, the merger is also modified to fall back on the default Pebble merger when the key is not of kind multihash-key. Further, the benchmarks are modified to explicitly assert that the inserted values are indeed found for each entry.
The
Getimplementation in Pebble store is also modified to copy the content returned by the backing store to avoid slice invalidation and runtime invalid reference errors captured in #80.Unrelated to the lookup issue, the changes here also fix typos in benchmarks where workload does not match the benchmark name. Further, the common tests are updated to do a flush as suggested in #93.
Fixes #94, #80.