Kadir OZDEMIR created PHOENIX-6150:
--------------------------------------
Summary: Repair ahead global indexes after concurrent mutations
Key: PHOENIX-6150
URL: https://issues.apache.org/jira/browse/PHOENIX-6150
Project: Phoenix
Issue Type: Improvement
Affects Versions: 4.15.0, 5.0.0
Reporter: Kadir OZDEMIR
Concurrent mutations processed by the IndexRegionObserver coproc leave (mutable
global) index rows in the unverified state and these unverified rows are
repaired by the GlobalIndexChecker coproc when they are read (i.e., scanned).
However, we do not have to wait for the scan operations to repair these
unverified rows. IndexRegionObserver can repair them asynchronously soon after
they written. There are two main reasons for proactively repairing index rows.
Efficient Repair: Repairing index rows within IndexRegionObserver eliminates
RPCs to read data table rows as IndexRegionObserver will read data table rows
locally.
Faster Scans: Read repair by GlobalIndexChecker is done one index row at a
time and each repair can take several tens of milliseconds. This adds
additional latency on the scan operations. Reducing the number of such read
repairs will improve scan performance.
This repair ahead can be optional and configurable. We can also have metrics
to monitor the number of index rows repaired ahead.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
