Note: I forget to record the meetings and the beginning (again), but
started the recording just before the beginning of the presentation
about the Upgrade feature (see bellow).
>>> Current state of the release / big features (Jitendra)
* ofs might require a few cleanup patches (but those can be done on
master branch) but the majority of work is done
* bucket bind mounts (required to make the s3 changes backward
compatible) patch can be expected in this week
>>> Merging Rocksdb on Datanode (on behalf of runzhiwang)
* runzhiwang created a proposal on the mailing list
* TLDR; thousands of containers on the datanode → thousands of
Rocksdb handlers → Memory / thread pressure
* Merge seems to be the easiest path
* Jitendra: before a bigger refactor, would be great to collect
more information about the overhead (how painful it is...)
>>> Balancing diskio between disks on datanode (Shashikant):
* https://issues.apache.org/jira/browse/HDDS-3700
* TLDR; If there are multiple disks (and/or multiraft is used) the
disks might not be utilized fully
* Proposal will be uploaded soon
>>> Bounded vs unbounded retry policy (Shashikant)
* https://issues.apache.org/jira/browse/HDDS-3611
* There is a generic question (bounded and unbounded retry) and a
very specific problem (Teragen 1T).
* Conclusion: would be great to have better understanding about the
problem
>>> Proposal for Upgrade
* Aravindan introduced the concept on high level
* TLDR;
* Layout versions are stored both in the release jars and the
metadata distributions and can be compared
* Binaries and metadata can be upgraded in two independent step
* (1. upgrade jars, 2. check everything works 3. enable new features)
RECORDING!!!!!
https://cloudera.zoom.us/rec/share/_ZRocInc_VNIa6vP6EGGfvE6MYe1X6a81XQX-6cOzByWIvvJXkML7uwKIrRup_ZP
Password: 3b%2263z
* Conclusion: questions raised about the technical details
(especially about rollback), we should continue the discussion (mailing
threads/folllow up meetings)
Marton
---------------------------------------------------------------------
To unsubscribe, e-mail: ozone-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: ozone-dev-h...@hadoop.apache.org
