[HBASE-15654] Optimize client's MetaCache handling

• Type: Umbrella
• Status: Resolved
• Priority: Critical
• Resolution: Implemented
• Affects Version/s: 1.3.0
• Component/s: Client
• Assignee: Mikhail Antonov
• Reporter: Mikhail Antonov
• Created: 14/Apr/16 18:33
• Updated: 11/Jun/22 18:51
• Resolved: 11/Jun/22 18:51

Description

This is an umbrella jira to track all individual issues, bugfixes and small optimizations around MetaCache (region locations cache) in the client. Motivation is that under the load one could see a spikes in the number of requests going to meta - reaching tens of thousands requests per second.

That covers issues when we clear entries from location cache unnecessary, as well as when we do more lookups than necessary when entries are legitimately evicted.

Issue Links

• contains: HBASE-20697 Can't cache All region locations of the specify table by calling table.getRegionLocator().getAllRegionLocations()

Sub-Tasks

• Use same criteria for clearing meta cache for all operations: Resolved
• Reduce number of concurrent region location lookups when MetaCache entry is cleared: Resolved
• Fix NPE in client-side metrics: Resolved
• Unnecessary MetaCache evictions cause elevated number of requests to meta: Resolved
• RegionServerCallable / RpcRetryingCaller clear meta cache on retries: Resolved
• HBase Client region location is expensive: Resolved
• FailedServerException shouldn't clear MetaCache: Resolved
• Normalize RetryingCallable cache clearing and implementations: Resolved