[
https://issues.apache.org/jira/browse/CASSANDRA-20209?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stefan Miklosovic updated CASSANDRA-20209:
------------------------------------------
Reviewers: Maxim Muzafarov
Status: Review In Progress (was: Needs Committer)
> Two snapshot manifests on same table against same SSTables do not have always
> same size
> ---------------------------------------------------------------------------------------
>
> Key: CASSANDRA-20209
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20209
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Local/Snapshots
> Reporter: Stefan Miklosovic
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 5.x
>
>
> I noticed that two manifests can have different sizes when snapshot is
> created on same SSTables. This might be hit in practice in tests when I
> measure that two snapshots have same sizes on disk.
> The difference is caused by manifest.json, check these two:
> {code}
> {
> "files" : [ "oa-1-big-Data.db" ],
> "created_at" : "2025-01-15T13:26:45.04Z",
> "expires_at" : null,
> "ephemeral" : false
> }
> {code}
> {code}
> {
> "files" : [ "oa-1-big-Data.db" ],
> "created_at" : "2025-01-15T13:27:45.012Z",
> "expires_at" : null,
> "ephemeral" : false
> }
> {code}
> The difference is in "created_at". One manifest has two decimals for
> milliseconds (04) and another one three (012).
> The difference is one digit, which will make the whole file bigger, hence
> their sizes do not match, hence the sizes of the snapshots as such do not
> match.
> This causes errors when e.g. running multiplex tests in CI. If test happens
> to generate two manifests with three places for milliseconds (or two) in both
> cases, then it will just pass but every now and then, it doesn't.
> The solution is to modify JsonUtils and persuade JSON_OBJECT_MAPPER to
> serialize Instant's always with 3 decimal places.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
