Chinmay Kulkarni created PHOENIX-6126:
-----------------------------------------
Summary: All createViewAddChildLink requests will go to the same
region of SYSTEM.CHILD_LINK
Key: PHOENIX-6126
URL: https://issues.apache.org/jira/browse/PHOENIX-6126
Project: Phoenix
Issue Type: Improvement
Affects Versions: 4.15.0, 5.0.0
Reporter: Chinmay Kulkarni
Fix For: 5.1.0, 4.16.0
Currently, since there is no split policy defined for SYSTEM.CHILD_LINK, it can
split by default.
If SYSTEM.CHILD_LINK does split, then when the client invokes
ConnectionQueryServices.childLinkMetaDataCoprocessorExec(), we always invoke
the endpoint on the first region that contains linking rows for that parent
(see
[this|https://github.com/apache/phoenix/blob/e3c7b4bdce2524eb4fd1e7eb0ccd3454fcca81ce/phoenix-core/src/main/java/org/apache/phoenix/query/ConnectionQueryServicesImpl.java#L1629]).
Thus, no matter which region would actually contain the parent→child link, the
first region is the one that has to always service the request and its RS
finally issues the mutation to the correct region. Note that the
mutateRowsWithLocks() call inside
[ChildLinkMetaDataEndpoint.createViewAddChildLink()|https://github.com/apache/phoenix/blob/e3c7b4bdce2524eb4fd1e7eb0ccd3454fcca81ce/phoenix-core/src/main/java/org/apache/phoenix/coprocessor/ChildLinkMetaDataEndpoint.java#L105]
doesn’t fail even if the region doesn’t contain that row since we pass in an
empty list for rowsToLock.
During view creation, this request will lead to a Put of just 1 row and we have
all the information to be able to pass in the exact row key instead of just the
prefix (parent tenantId, parent schema, parent name). This should distribute
load to all regions of SYSTEM.CHILD_LINK.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
