Resolves #860: Try planning an IN predicate as an OR of equalities. #861
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
|
It occurred to me that I should probably catch the |
MMcM
requested changes
Mar 19, 2020
...rd-layer-core/src/main/java/com/apple/foundationdb/record/query/plan/RecordQueryPlanner.java
Show resolved
Hide resolved
|
Re: DNF explosion, it looks like this part of the planner doesn't use the full normalization logic, although I'm not sure why. I added a test with a very large CNF (large enough to cause a stack overflow) to make sure this doesn't change. |
nschiefer
force-pushed
the
plan-in-as-or
branch
2 times, most recently
from
f1c7399
to
4ce8f79
Compare
nschiefer
commented
Mar 25, 2020
...rd-layer-core/src/main/java/com/apple/foundationdb/record/query/plan/RecordQueryPlanner.java
Outdated
Show resolved
Hide resolved
nschiefer
force-pushed
the
plan-in-as-or
branch
from
4ce8f79
to
4791a9d
Compare
MMcM
approved these changes
Mar 25, 2020
…qualities. If the planner detects that it can't implement an IN predicate using an in-join (for example, because of an incompatible sort order), it tries to rewrite it as an OR predicate instead. This attempt is controlled by a configuration object, which also includes the IndexScanPreference.
nschiefer
force-pushed
the
plan-in-as-or
branch
from
4791a9d
to
0546106
Compare
alecgrieser
requested changes
Mar 25, 2020
...rd-layer-core/src/main/java/com/apple/foundationdb/record/query/plan/RecordQueryPlanner.java
Outdated
Show resolved
Hide resolved
.../src/main/java/com/apple/foundationdb/record/query/plan/RecordQueryPlannerConfiguration.java
Show resolved
Hide resolved
...rd-layer-core/src/main/java/com/apple/foundationdb/record/query/plan/RecordQueryPlanner.java
Show resolved
Hide resolved
...-layer-core/src/main/java/com/apple/foundationdb/record/query/plan/planning/InExtractor.java
Show resolved
Hide resolved
|
|
||
| // If we don't change this when shouldAttemptFailedInJoinAsOr() is true, then we _always_ pick the union plan, | ||
| // rather than the in join plan. | ||
| int score = getConfiguration().shouldAttemptFailedInJoinAsOr() ? 0 : 1; |
There was a problem hiding this comment.
Hmm, interesting. Should the logic in the join filter that chooses between the Union and the InJoin be changed then (maybe ignoring the score if we can't trust it?).
There was a problem hiding this comment.
Possibly. I'm a bit worried about doing so because I don't really know what the score is trying to capture. If we think that's lower risk, I'm for it.
.../src/test/java/com/apple/foundationdb/record/provider/foundationdb/query/FDBInQueryTest.java
Outdated
Show resolved
Hide resolved
.../src/test/java/com/apple/foundationdb/record/provider/foundationdb/query/FDBInQueryTest.java
Outdated
Show resolved
Hide resolved
.../test/java/com/apple/foundationdb/record/query/plan/RecordQueryPlannerConfigurationTest.java
Outdated
Show resolved
Hide resolved
alecgrieser
approved these changes
Mar 25, 2020
nschiefer
added a commit
that referenced
this pull request
Apr 3, 2020
PR #861 added a new transformation to the RecordQueryPlanner that attempts to transform an IN predicate into an equivalent OR of equality predicates when the planner determines that it cannot implement the IN predicate as an IN-join because of an incompatible sort order. However, the equivalent OR predicate is too complex for the simple normalizer normalizeAndOr() if there are other predicates on the original filter. In this case, the predicates are not normalized and so the union planner is unable to produce the desired union plan. This change adds a more sophisticated normalizer to handle this case.
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.
If the planner detects that it can't implement an IN predicate using an in-join (for example, because of an incompatible sort order), it tries to rewrite it as an OR predicate instead.
See #860 for a description of the specific problem that this solves.
The implementation is a bit hacky and ends up relying on pieces of the new planner to make decisions. This might not be great hygiene, but it seemed like the right tool for determining when to use one plan or another.