[HIVE-16520] Cache hive metadata in metastore

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0
• Component/s: Metastore
• Target Version/s: 3.0.0
• Hadoop Flags: Reviewed
• Release Note: To use CachedStore, please set hive.metastore.rawstore.impl to "org.apache.hadoop.hive.metastore.cache.CachedStore" in hive-site.xml.
• Assignee: Daniel Dai
• Reporter: Daniel Dai
• Created: 24/Apr/17 18:04
• Updated: 01/Nov/18 19:33
• Resolved: 01/May/17 20:55

Description

During Hive 2 benchmark, we find Hive metastore operation take a lot of time and thus slow down Hive compilation. In some extreme case, it takes much longer than the actual query run time. Especially, we find the latency of cloud db is very high and 90% of total query runtime is waiting for metastore SQL database operations. Based on this observation, the metastore operation performance will be greatly enhanced if we have a memory structure which cache the database query result.

Issue Links

• is related to: HIVE-18039 Use ConcurrentHashMap for CachedStore
• links to: GitHub Pull Request #173
• links to: Proposal

Sub-Tasks

• Fix Unit test failures when CachedStore is enabled: Closed
• CachedStore: improvements to partition col stats caching and cache column stats for unpartitioned table: Closed
• CachedStore: make prewarm and background cache update multithreaded: Open
• Fix remaining unit test failures when CachedStore is enabled: Open
• CachedStore: Store cached partitions/col stats within the table cache and make prewarm non-blocking: Resolved
• CachedStore: Prioritize loading of recently accessed tables during prewarm: Closed
• CachedStore: Investigate TestCachedStore#testTableColStatsOps: Open
• CachedStore: Use metastore notification log events to update cache: Resolved
• CachedStore leak PersistenceManager resources: Closed
• NPE during CachedStore refresh: Closed
• CachedStore - wait for prewarm at use time, not init time: Closed
• CachedStore: Have a whitelist/blacklist config to allow selective caching of tables/partitions and allow read while prewarming: Closed
• CachedStore: prewarm improvement (avoid multiple sql calls to read partition column stats), refactoring and caching some aggregate stats: Closed
• CachedStore: Use memory estimation to limit cache size during prewarm: Closed
• CachedStore: bug fixes for TestEmbeddedHiveMetaStore, TestRemoteHiveMetaStore, TestMiniLlapCliDriver, TestMiniTezCliDriver, TestMinimrCliDriver: Resolved
• CachedStore: bug fixes for q file tests: TestMiniLlapCliDriver, TestMiniTezCliDriver, TestMinimrCliDriver: Resolved
• CachedStore: Fix UT when CachedStore is enabled: Patch Available
• CachedStore: Add more UT coverage (outside of .q files): Closed
• CachedStore: Run a select q file tests with CachedStore enabled: Open
• CachedStore: Background refresh thread bug fixes: Closed