Ivan Bessonov created IGNITE-26305:
--------------------------------------
Summary: Optimize log scan
Key: IGNITE-26305
URL: https://issues.apache.org/jira/browse/IGNITE-26305
Project: Ignite
Issue Type: Improvement
Reporter: Ivan Bessonov
During the local recovery and log replication we need to read multiple log
entries in order. Performing a full lookup in-memory metadata lookup takes too
much time ({{{}O(log){}}}) and should be avoided if possible.
We should either develop a prefetch, or implement a thread-local state of last
read in-memory metadata block and continue iterating over it. All we need to
care about here is a truncated suffix, or other types of meta-data that can be
invalidated (of course, it's not that simple).
In other words, using a benchmark from
https://issues.apache.org/jira/browse/IGNITE-26293, or something similar, we
should take a look at bottlenecks and optimize them if possible.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
