Ivan Bessonov created IGNITE-26281:
--------------------------------------
Summary: Implement in-memory metadata storage for
Key: IGNITE-26281
URL: https://issues.apache.org/jira/browse/IGNITE-26281
Project: Ignite
Issue Type: Improvement
Reporter: Ivan Bessonov
What is it: it's an in-memory representation of already persisted metadata for
index files described here: https://issues.apache.org/jira/browse/IGNITE-26279.
Schematically it should look like this:
{code:java}
InMemoryMetadata = Map<
ReplicationGroupId,
GroupMetadata
>
GroupMetadata {
Queue<BlockOfIndexMetas>
}
BlockOfIndexMetas = {
startRaftIndex
endRaftIndex
startIndexId
IndexMeta[]
}
IndexMeta = {
startRaftIndex
generation
offsetInIndexFile
} {code}
{{InMemoryMetadata}} will also be spread across individual log storages of
individual raft groups for fast access.
{{GroupMetadata}} represents a queue of index files, so to speak, from newest
to oldest. It reflects the lookup order in the queue when we're executing the
{{getEntry}} operation.
{{IndexMeta[]}} groups together all index files for which the {{End}} of a
previous meta matches {{Start}} of next meta. This means that a particularly
large {{truncateSuffix}} might lead to adding a new {{BlockOfIndexMetas}}
instance to the queue.
In this Jira we should maintain this structure from the checkpoint thread. We
should try to make it block-free, so that other threads can read from it while
we're writing.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
