[
https://issues.apache.org/jira/browse/JAMES-3937?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benoit Tellier updated JAMES-3937:
----------------------------------
Description:
h3. Why?
CassandraThreadIdGuessingAlgorithm tables occupies a non neglictible amount of
space.
Out of a 20 GB database I have in one of my production platform:
{code:java}
Table: threadlookuptable
SSTable count: 4
Space used (total): 360 263 739
Table: threadtable
SSTable count: 8
Space used (total): 1 050 590 715
{code}
Which is non neglictible.
The goal here would be to reduce the space used in database by thread
allocation.
h4. Other concerns
Storing subjects as is linked to usernames is likely problematic in terms of
privacy.
h3. How ?
Thread Guessing Algorithm do not need raw values to operate but works with
hashs as demonstrated by https://issues.apache.org/jira/browse/JAMES-3937.
h3. Impact ?
As threads are partitionned by users risk of collision is extremly low and
false posotives might only result in incorrect thread grouping, making this use
case none sensitive to hash collisions. Use of non cryptographic hash methods
is thus acceptable.
We expect a significant space reduction.
*Migration*: We will just create a new table and drop the old one. THis will
cause a discontinuity in thread allocation: 2 threads instead of one. This
seems acceptable and preferable to a complex migration in our eyes.
was:
h3. Why?
CassandraThreadIdGuessingAlgorithm tables occupies a non neglictible amount of
space.
Out of a 20 GB database I have in one of my production platform:
{code:java}
Table: threadlookuptable
SSTable count: 4
Space used (total): 360 263 739
Table: threadtable
SSTable count: 8
Space used (total): 1 050 590 715
{code}
Which is non neglictible.
The goal here would be to reduce the space used in database by thread
allocation.
h4. Other concerns
Storing subjects as is linked to usernames is likely problematic in terms of
privacy.
h3. How ?
Thread Guessing Algorithm do not need raw values to operate but works with
hashs as demonstrated by [insert poc PR link here].
h3. Impact ?
As threads are partitionned by users risk of collision is extremly low and
false posotives might only result in incorrect thread grouping, making this use
case none sensitive to hash collisions. Use of non cryptographic hash methods
is thus acceptable.
We expect a significant space reduction.
*Migration*: We will just create a new table and drop the old one. THis will
cause a discontinuity in thread allocation: 2 threads instead of one. This
seems acceptable and preferable to a complex migration in our eyes.
> Improve CassandraThreadIdGuessingAlgorithm
> ------------------------------------------
>
> Key: JAMES-3937
> URL: https://issues.apache.org/jira/browse/JAMES-3937
> Project: James Server
> Issue Type: Improvement
> Components: cassandra, mailbox
> Affects Versions: 3.8.0
> Reporter: Benoit Tellier
> Priority: Major
> Fix For: 3.9.0
>
>
> h3. Why?
> CassandraThreadIdGuessingAlgorithm tables occupies a non neglictible amount
> of space.
> Out of a 20 GB database I have in one of my production platform:
> {code:java}
> Table: threadlookuptable
> SSTable count: 4
> Space used (total): 360 263 739
> Table: threadtable
> SSTable count: 8
> Space used (total): 1 050 590 715
> {code}
> Which is non neglictible.
> The goal here would be to reduce the space used in database by thread
> allocation.
> h4. Other concerns
> Storing subjects as is linked to usernames is likely problematic in terms of
> privacy.
> h3. How ?
> Thread Guessing Algorithm do not need raw values to operate but works with
> hashs as demonstrated by https://issues.apache.org/jira/browse/JAMES-3937.
> h3. Impact ?
> As threads are partitionned by users risk of collision is extremly low and
> false posotives might only result in incorrect thread grouping, making this
> use case none sensitive to hash collisions. Use of non cryptographic hash
> methods is thus acceptable.
> We expect a significant space reduction.
> *Migration*: We will just create a new table and drop the old one. THis will
> cause a discontinuity in thread allocation: 2 threads instead of one. This
> seems acceptable and preferable to a complex migration in our eyes.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
