Benoit Tellier created JAMES-4223:
-------------------------------------
Summary: S3: Leverage optionally If-None-Match header
Key: JAMES-4223
URL: https://issues.apache.org/jira/browse/JAMES-4223
Project: James Server
Issue Type: Improvement
Components: s3
Reporter: Benoit Tellier
h3. Why?
With deduplication enabled, James does 1 PUT each time it stores a given mail:
- That's innefficient as the underlying S3 would overwrite its content...
- That's harmfull with versionning on as each store would result in a new
version
h3. What?
Have a ` objectstorage.s3.ifNoneMatch.enable` option in blob.properties to set
`If-None-Match: *` headers on PUT.
This enforce S3 never overwrite an object and instead returns a 412
Precondition fail that we'd consider a success.
h3. Consequences
Ability to create a Contingency Plan without versions:
- s3:if-none-match policy rejects PUT without the If-None-Match header
enforcing content immutability (in your head, crypto lockers)
- s3:DeleteObject enforces no deletions and GC / deletes are done in a
non-internet exposed process.
h3. Alternative?
HEAD before PUT had been disregarded for performance reasons: it puts double
the amount of work on the S3 metadata
LINAGORA have a Cassandra stored blobId list extension not to do the same PUT
twice, thought as a performance optimization in its Twake Mail product. While
it works great it's a perf optimisation and consistency guaranty can not
enforce that we wont be doing the same PUT twice. blobIdList is tiered but it
can be coupled with GC generation however upon C* errors we might duplicate
PUT. Cassandra storage dedicated to this is non neglictible. My though is that
the two could be complementary: blobIdList empowers skipping content streaming.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
