Description
Recently we switched an MR application from scanning live tables to scanning snapshots (PHOENIX-3744). We ran into a severe performance issue, which turned out to a correctness issue due to over-lapping scan splits generation. After some debugging we figured that it has been fixed via PHOENIX-4997.
We also need not restore the snapshot per map task. Currently, we restore the snapshot once per map task into a temp directory. For large tables on big clusters, this creates a storm of NN RPCs. We can do this once per job and let all the map tasks operate on the same restored snapshot. HBase already did this via HBASE-18806, we can do something similar. Jira to correct this behavior: https://issues.apache.org/jira/browse/PHOENIX-6334
The purpose of this Jira is to resolve this issue immediately by providing the ability to the caller to decide whether or not snapshot restore needs to be handled externally or internally on the Phoenix side (the buggy approach).
All other performance suggestions here: https://issues.apache.org/jira/browse/PHOENIX-6081
Attachments
Issue Links
- relates to
-
PHOENIX-6081 Improvements to snapshot based MR input format
- Open
- links to