Benoit Tellier created JAMES-3458:
-------------------------------------
Summary: [IMAP] GetQuotaRoot commands is issuing more than 10% of
cassandra qurey time!
Key: JAMES-3458
URL: https://issues.apache.org/jira/browse/JAMES-3458
Project: James Server
Issue Type: Improvement
Components: cassandra, IMAPServer
Reporter: Benoit Tellier
Fix For: 3.6.0
Attachments: Capture d’écran de 2020-11-27 10-58-05.png, Capture
d’écran de 2020-11-27 10-58-28.png
# Why
IMAP clients performs a GETQUOTAROOT command for each mailboxes.
Untagged responses also include the actual quota; we end up because of IMAP
design reading the quotas for each mailboxes! This is an issue eg when a given
user have hundreds of mailboxes! (Who said IMAP is a BAD protocol?)
# How
While patching IMAP design is not doable, we can optimize our queries to reduce
our query count: storage and count quota parts (current usage, max user
storage, max domain storage) are hosted on the same row, that we are reading
two times due to bad readpath design.
Unlocking grouped read for collocated data would likely make this much less of
an issue.
**Definition of done**: GetQuotaRoot cassandra query volume should be reduced
of 40%
Glowroot captures attached. The first shows queries executed by GetQuotaroot,
as well as the percentage of total IMAP time spent on GetQuotaRoot queries
resolution.
As a comparison the second capture shows all the IMAP queries being issued...
# Impact
Note that this work would also enhance slightly append speed (SMTP, IMAP
append, etc) as quota needs to be checked, backend quota processing (quota
updated events are less expensive to generate), and will slightly help
regarding JMAP Mailbox/get call (which reads the quota). Though the enhancement
would be less noticeable than for GetQuotaRoot IMAP command.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
