Optimize the location overlap check with an index #1686

eric-maynard · 2025-05-27T23:34:16Z

The location overlap check for "sibling" tables (those which share a parent) has been a performance bottleneck since its introduction, but we haven't historically had a good way around this other than just disabling the check.

Current Behavior

The current logic is that when we create a table, we list all sibling tables and check each and every one to ensure there is no location overlap. This results in O(N^2) checks when adding N tables to a namespace, quickly becoming untenable.

With the CreateTreeDataset benchmark I tested creating 5000 sibling tables using the current code:

It is apparent that latency increases over time. Runs took between 90 and 200+ seconds, and Polaris instances with a small memory allocation were prone to crashing due to OOMs:

Proposed change

This PR adds a new persistence API, hasOverlappingSiblings, which if implemented can be used to directly check for the presence of siblings at the metastore layer.

This API is implemented for the JDBC metastore in a new schema version, and some changes are made to account for an evolving schema version now and in the future.

This implementation breaks a location down into components and queries for a sibling at each of those locations, so a new table at location s3://bucket/root/n1/nA/t1/ will require checking for an entity with location s3://bucket/, s3://bucket/root/, s3://bucket/root/n1/, s3://bucket/root/n1/nA/, and finally s3://bucket/root/n1/nA/t1/%. All of this can be done in a single query which makes a single pass over the data.

The query is optimized by the introduction of a new index over a new location column.

With the changes enabled, I tested creating 5000 sibling tables:

Latency is stable over time, and runs consistently completed in less than 30 seconds. I did not observe any OOMs when testing with the feature enabled.

…heck

singhpk234

Other round of feedbacks, apologies these feedbacks are not strictly on the changed diff, i missed reviewing some parts

persistence/relational-jdbc/src/main/resources/h2/schema-v2.sql

...al-jdbc/src/test/java/org/apache/polaris/persistence/relational/jdbc/QueryGeneratorTest.java

polaris-core/src/main/java/org/apache/polaris/core/persistence/BasePersistence.java

...tional-jdbc/src/main/java/org/apache/polaris/persistence/relational/jdbc/QueryGenerator.java

...-jdbc/src/main/java/org/apache/polaris/persistence/relational/jdbc/models/SchemaVersion.java

polaris-core/src/main/java/org/apache/polaris/core/entity/LocationBasedEntity.java

…heck

...tional-jdbc/src/main/java/org/apache/polaris/persistence/relational/jdbc/QueryGenerator.java

persistence/relational-jdbc/src/main/resources/h2/schema-v2.sql

...bc/src/main/java/org/apache/polaris/persistence/relational/jdbc/JdbcBasePersistenceImpl.java

singhpk234 · 2025-06-24T23:29:50Z

...lational-jdbc/src/main/java/org/apache/polaris/persistence/relational/jdbc/DatabaseType.java


  public String getInitScriptResource() {
-    return String.format("%s/schema-v1.sql", this.getDisplayName());
+    return String.format("%s/schema-v2.sql", this.getDisplayName());


wouldn't version be passed here ? as it will start bootstraping always by v2 ?

Isn't it good to bootstrap with the latest version? Do we need to wait until users can select a schema version to bootstrap with? Currently it's hardcoded.

Thats true v1 is the only version though 🤔, can't they make a choice ? wdyt ?

I think that's probably a separate feature -- in theory, it would be good to let users provide their own .sql file to bootstrap from. I can pick this up separately if possible? Actually, I already spent some time trying this, but I ran into some issues with the Quarkus datasource setup.

sounds good, i think may be taking version at the time of bootstrapping in from the command line and then percolating all the way through could also be an option ? this way the database type is still inferred using the the existing mechanism

Good idea, let's do that... check out #1942 for a continuation of this!

...bc/src/main/java/org/apache/polaris/persistence/relational/jdbc/JdbcBasePersistenceImpl.java

...tional-jdbc/src/main/java/org/apache/polaris/persistence/relational/jdbc/QueryGenerator.java

polaris-core/src/main/java/org/apache/polaris/core/entity/LocationBasedEntity.java

polaris-core/src/main/java/org/apache/polaris/core/persistence/BasePersistence.java

singhpk234

LGTM, thanks @eric-maynard ! I don't have any comments, thank you so much for working on version aware bootstraping

Instead of always using the hardcoded `schema-v1.sql` file, it would be nice if users could specify a file to bootstrap from. This is especially relevant after #1686 which proposes to add a new "version" of the schema.

The location overlap check for "sibling" tables (those which share a parent) has been a performance bottleneck since its introduction, but we haven't historically had a good way around this other than just disabling the check. <hr> ### Current Behavior The current logic is that when we create a table, we list all sibling tables and check each and every one to ensure there is no location overlap. This results in O(N^2) checks when adding N tables to a namespace, quickly becoming untenable. With the `CreateTreeDataset` [benchmark](https://github.com/eric-maynard/polaris-tools/blob/main/benchmarks/src/gatling/scala/org/apache/polaris/benchmarks/simulations/CreateTreeDataset.scala) I tested creating 5000 sibling tables using the current code: It is apparent that latency increases over time. Runs took between 90 and 200+ seconds, and Polaris instances with a small memory allocation were prone to crashing due to OOMs: ### Proposed change This PR adds a new persistence API, `hasOverlappingSiblings`, which if implemented can be used to directly check for the presence of siblings at the metastore layer. This API is implemented for the JDBC metastore in a new schema version, and some changes are made to account for an evolving schema version now and in the future. This implementation breaks a location down into components and queries for a sibling at each of those locations, so a new table at location `s3://bucket/root/n1/nA/t1/` will require checking for an entity with location `s3://bucket/`, `s3://bucket/root/`, `s3://bucket/root/n1/`, `s3://bucket/root/n1/nA/`, and finally `s3://bucket/root/n1/nA/t1/%`. All of this can be done in a single query which makes a single pass over the data. The query is optimized by the introduction of a new index over a new _location_ column. With the changes enabled, I tested creating 5000 sibling tables: Latency is stable over time, and runs consistently completed in less than 30 seconds. I did not observe any OOMs when testing with the feature enabled. (cherry picked from commit 1e69373)

* Exclude unused dependency for polaris spark client dependency (apache#1933) * enable ETag integration tests (apache#1935) tests were added in 8b5dfa9 and afaict supposed to get enabled after ec97c1b * Fix Pagination for Catalog Federation (apache#1849) Details can be found in this issue: apache#1848 * Update doc to fix docker build inconsistency issue (apache#1946) * Simplify install dependency doc (apache#1941) * Simply getting start doc * Simply install dependecy doc * Minor words change * Fix admin tool for quick start (apache#1945) When attempting to use the `polaris-admin-tool.jar` to bootstrap a realm, the application fails with a `jakarta.enterprise.inject.UnsatisfiedResolutionException` because it cannot find a `javax.sql.DataSource` bean. Detail in apache#1943 This issue occurs because `quarkus.datasource.db-kind`is a build-time property in Quarkus. Its value must be defined during the application's build process to enable the datasource extension and generate the necessary CDI bean producer (ref: https://quarkus.io/guides/all-config#quarkus-datasource_quarkus-datasource-db-kind). I think we only support postgres for now, thus, I set `quarkus.datasource.db-kind=postgresql`. This can be problematic if we later want to support more data sources other than postgres. There are couple options we have for this such as use multiple named datasources in the config during build time. But this may be out of scope of this PR. I am open for more discussion on this, but for the time being, it may be better to unblock people who are trying to use the quick start doc. Sample output for the bootstrap container after the fix: ``` ➜ polaris git:(1943) docker logs polaris-polaris-bootstrap-1 Realm 'POLARIS' successfully bootstrapped. Bootstrap completed successfully. ``` * fix(build): Fix deprecation warnings in FeatureConfiguration (apache#1894) * Fix NPE in listCatalogs (apache#1949) listCatalogs is non-atomic. It first atomically lists all entities and then iterates through each one and does an individual loadEntity call. This causes an NPE when calling `CatalogEntity::new`. I don't think it's ever useful for listCatalogsUnsafe to return null since the caller isn't expecting a certain length of elements, so I just filtered it there. * Fix doc for sample log and default password (apache#1951) Minor updates for the quick start doc: 1. update sample output to reflect with the latest code 2. update default password to the right value 3. remove trailing space * Optimize the location overlap check with an index (apache#1686) The location overlap check for "sibling" tables (those which share a parent) has been a performance bottleneck since its introduction, but we haven't historically had a good way around this other than just disabling the check. <hr> ### Current Behavior The current logic is that when we create a table, we list all sibling tables and check each and every one to ensure there is no location overlap. This results in O(N^2) checks when adding N tables to a namespace, quickly becoming untenable. With the `CreateTreeDataset` [benchmark](https://github.com/eric-maynard/polaris-tools/blob/main/benchmarks/src/gatling/scala/org/apache/polaris/benchmarks/simulations/CreateTreeDataset.scala) I tested creating 5000 sibling tables using the current code: It is apparent that latency increases over time. Runs took between 90 and 200+ seconds, and Polaris instances with a small memory allocation were prone to crashing due to OOMs: ### Proposed change This PR adds a new persistence API, `hasOverlappingSiblings`, which if implemented can be used to directly check for the presence of siblings at the metastore layer. This API is implemented for the JDBC metastore in a new schema version, and some changes are made to account for an evolving schema version now and in the future. This implementation breaks a location down into components and queries for a sibling at each of those locations, so a new table at location `s3://bucket/root/n1/nA/t1/` will require checking for an entity with location `s3://bucket/`, `s3://bucket/root/`, `s3://bucket/root/n1/`, `s3://bucket/root/n1/nA/`, and finally `s3://bucket/root/n1/nA/t1/%`. All of this can be done in a single query which makes a single pass over the data. The query is optimized by the introduction of a new index over a new _location_ column. With the changes enabled, I tested creating 5000 sibling tables: Latency is stable over time, and runs consistently completed in less than 30 seconds. I did not observe any OOMs when testing with the feature enabled. * Add SUPPORTED_EXTERNAL_CATALOG_AUTHENTICATION_TYPES feature configuration (apache#1931) * Add SUPPORTED_FEDERATION_AUTHENTICATION_TYPES feature configuration * Add unit tests * Update Helm chart version (apache#1957) * Remove the maintainer list in Helm Chart README (apache#1962) * Use multi-lines instead of single line (apache#1961) * Fix invalid sample script in CLI doc (apache#1964) * Fix hugo blockquote (apache#1967) * Fix hugo blockquote * Add license header * Fix lint rules (apache#1953) * Mutable objects used for immutable values (apache#1596) * fix: Only include project LICENSE and NOTICE in Spark Client Jar (apache#1950) * Add Sushant as a collaborator (apache#1956) * Adds missing Google Flatbuffers license information (apache#1968) * fix: Typo in Spark Client Build File (apache#1969) debugrmation * Python code format (apache#1954) * test(integration): refactor PolarisRestCatalogIntegrationTest to run against any cloud provider (apache#1934) * Make Catalog Integration Test suite cloud native * Fix admin tool doc (apache#1977) * Fix admin tool doc * Fix admin tool doc * Update release-guide.md (apache#1927) * Add relational-jdbc to helm (apache#1937) Motivation for the Change Polaris needs to support relational-jdbc as the default persistence type for simpler database configuration and better cloud-native deployment experience. Description of the Status Quo (Current Behavior) Currently, the Helm chart only supports eclipse-link persistence type as the default, which requires complex JPA configuration with persistence.xml files. Desired Behavior Add relational-jdbc persistence type support to Helm chart Use relational-jdbc as the default persistence type Inject JDBC configuration (username, password, jdbc_url) through Kubernetes Secrets as environment variables Maintain backward compatibility with eclipse-link Additional Details Updated persistence-values.yaml for CI testing Updated test coverage for relational-jdbc configuration JDBC credentials are injected via QUARKUS_DATASOURCE_* environment variables from Secret Secret keys: username, password, jdbc_url * Add CHANGELOG (apache#1952) * Add rudimentary CHANGELOG.md * Add the Jetbrains Changelog Gradle plugin to help managing CHANGELOG.md * Share Polaris Community Meeting for 2025-06-26 (apache#1978) * Correct javadoc text in generateOverlapQuery() (apache#1975) * Fix javadoc warning: invalid input: '&' * Correct javadoc text in generateOverlapQuery() * Do not serialize null properties in the management model (apache#1955) * Ignore null values in JSON output * This may have an impact on existing client, but it is not likely to be substantial because normally absent properties should be treated the same as having `null` values. * This change enables adding new optional fields to the Management API while maintaining backward compatibility in the future: New properties will not be exposed to clients unless a value for them in explicitly set. * Add OpenHFT in Spark plugin LICENSE (apache#1979) * Add additional unit and integration tests for etag functionality (apache#1972) * Additional unit test for Etags * Added a few corner case IT tests for testing etags with schema changes. * Added IT tests to test changes after DDL and DML * Add options to the bootstrap command to specify a schema file (apache#1942) Instead of always using the hardcoded `schema-v1.sql` file, it would be nice if users could specify a file to bootstrap from. This is especially relevant after apache#1686 which proposes to add a new "version" of the schema. * Added support for `s3a` scheme (apache#1932) * Fix the sign failure (apache#1926) * Fix doc to remove outdated note about fine-grained access controls support (apache#1983) Minor update for the access control doc: 1. Remove the misleading section on privileges can only be granted at catalog level. I've tested the fine-grained access controls and confirmed that privileges can be applied to an individual table in the catalog. * Add support for catalog federation in the CLI (apache#1912) The CLI currently only supports the version of EXTERNAL catalogs that was present in 0.9.0. Now, EXTERNAL catalogs can be configured with various configurations relating to federation. This PR updates the CLI to better match the REST API so that federated catalogs can be easily set up in the CLI. * fix: Remove db-kind in helm chart (apache#1987) * Add a Spark session builder for the tests (apache#1985) * Fix doc for CLI update (apache#1994) PR for apache#1866 * Improve createPrincipal example in API docs (apache#1992) In apache#1929 it was pointed out that the example in the Polaris docs suggests that users can provide a client ID during principal creation: . . . This PR attempts to fix this by adding an explicit example to the spec. * Add doc for repair option (apache#1993) PR for apache#1864 * Refactor relationalJdbc in helm (apache#1996) * Add regression test coverage for Spark Client with package conf (apache#1997) * Remove unnecessary `InputStream.close` call (apache#1982) apache#1942 changed the way that the bootstrap init script is handled, but it added an extra `InputStream.close` call that shouldn't be needed after the BufferedReader [here](https://github.com/apache/polaris/pull/1942/files#diff-de43b240b5b5e07aba7e89f5515a417cefd908845b85432f3fcc0819911f3e2eR89) is closed. This PR removes that extra call. * Materialize Realm ID for Session Supplier in JDBC (apache#1988) It was discovered that the Session Supplier maps used in the MetaStoreManagerFactory implementations were passing in RealmContext objects to the supplier directly and then using the RealmContext objects to create BasePersistence implementation objects within the supplier. This supplier is cached on a per-realm basis in most MetaStoreManagerFactory implementations. RealmContext objects are request-scoped beans. As a result, if any work is being done outside the scope of the request, such as during a Task, any calls to getOrCreateSessionSupplier for creating a BasePersistence implementation will fail as the RealmContext object is no longer available. This PR will ensure for the JdbcMetaStoreManagerFactory that the Realm ID is materialized from the RealmContext and used inside the supplier so that the potentially deactivated RealmContext object does not need to be used in creating the BasePersistence object. Given that we are caching on a per-realm basis, this should not introduce any unforeseen behavior for the JdbcMetaStoreManagerFactory as the Realm ID must match exactly for the same supplier to be returned from the Session Supplier map. * rebase/changes * minor refactoring * Last merged commit 8fa6bf2 --------- Co-authored-by: Yun Zou <[email protected]> Co-authored-by: Christopher Lambert <[email protected]> Co-authored-by: Rulin Xing <[email protected]> Co-authored-by: MonkeyCanCode <[email protected]> Co-authored-by: Alexandre Dutra <[email protected]> Co-authored-by: Andrew Guterman <[email protected]> Co-authored-by: Eric Maynard <[email protected]> Co-authored-by: Pooja Nilangekar <[email protected]> Co-authored-by: Yufei Gu <[email protected]> Co-authored-by: fabio-rizzo-01 <[email protected]> Co-authored-by: Russell Spitzer <[email protected]> Co-authored-by: Sushant Raikar <[email protected]> Co-authored-by: Jiwon Park <[email protected]> Co-authored-by: Dmitri Bourlatchkov <[email protected]> Co-authored-by: JB Onofré <[email protected]> Co-authored-by: Sandhya Sundaresan <[email protected]> Co-authored-by: Pavan Lanka <[email protected]> Co-authored-by: CG <[email protected]> Co-authored-by: Adnan Hemani <[email protected]>

eric-maynard added 17 commits May 23, 2025 11:00

initial commit

ba0ec2b

add index

6ee1262

changes

5426f16

needs refactoring

2f8ad59

wire model

7172b05

continue wiring

8a6f567

feature configs

f8f615c

building

911b37b

testing

b731fe1

failed to find version

ef8e024

stable

79b8ec2

slash fix

70a3481

stable with slash

6529a84

testing

ac0b17e

Merge branch 'main' of github.com:apache/polaris into index-sibling-c…

d8a3140

…heck

sql fix

30412b5

review

7e69c97

eric-maynard requested review from MonkeyCanCode, RussellSpitzer, adutra, ashvina, collado-mike, dennishuo, dimas-b, flyrain, jackye1995, jbonofre, snazy, takidau and vvcephei as code owners May 27, 2025 23:34

eric-maynard added 4 commits June 16, 2025 15:14

add extra check

d14b8e2

autolint

acc2b09

normalization test case fixed up

5545ece

autolint

4edc032

singhpk234 reviewed Jun 16, 2025

View reviewed changes

eric-maynard mentioned this pull request Jun 17, 2025

Implement benchmark scenario WeightedWorkloadOnTreeDataset apache/polaris-tools#21

Merged

eric-maynard added 3 commits June 17, 2025 08:51

changes per review

609d997

Merge branch 'main' of github.com:apache/polaris into index-sibling-c…

9703a93

…heck

fix new issue related to withoutScheme

28056d4

singhpk234 reviewed Jun 20, 2025

View reviewed changes

eric-maynard added 2 commits June 19, 2025 22:38

missed a comment

1eeb522

resolve conflicts

735e9a9

eric-maynard requested a review from singhpk234 June 24, 2025 17:07

eric-maynard added 2 commits June 24, 2025 10:07

polish

9730cbf

autolint

f85642b

singhpk234 reviewed Jun 25, 2025

View reviewed changes

eric-maynard added 2 commits June 25, 2025 09:36

comment changes

f446486

autolint

e689d72

eric-maynard mentioned this pull request Jun 25, 2025

Add options to the bootstrap command to specify a schema file #1942

Merged

eric-maynard added 2 commits June 25, 2025 12:09

another change per review

2ff0e11

autolint

c6e5edb

singhpk234 approved these changes Jun 25, 2025

View reviewed changes

github-project-automation bot moved this from PRs In Progress to Ready to merge in Basic Kanban Board Jun 25, 2025

eric-maynard merged commit 1e69373 into apache:main Jun 26, 2025
11 checks passed

github-project-automation bot moved this from Ready to merge to Done in Basic Kanban Board Jun 26, 2025

eric-maynard mentioned this pull request Jun 26, 2025

Fixed #214 - Added support for s3a scheme #1932

Merged

HonahX mentioned this pull request Jan 7, 2026

Table Creation Fails with 500 when OPTIMIZED_SIBLING_CHECK=true with JDBC Persistence #3378

Open

Optimize the location overlap check with an index #1686

Optimize the location overlap check with an index #1686

Uh oh!

Conversation

eric-maynard commented May 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Current Behavior

Proposed change

Uh oh!

singhpk234 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

singhpk234 Jun 24, 2025

Choose a reason for hiding this comment

Uh oh!

eric-maynard Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

singhpk234 Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

eric-maynard Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

singhpk234 Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

eric-maynard Jun 25, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

singhpk234 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

eric-maynard commented May 27, 2025 •

edited

Loading