| timburke_ | mattoliver, cool! the test failures have me a little worried, though -- might have something to do with some of the naming decisions i made at the head of the chain? | 00:03 |
|---|---|---|
| timburke_ | i'm not sure i like how RingSerializeBase et al. spread the v2 codec logic out so much... i think it's making some of the errors i'm seeing harder to reason about... `AttributeError: can't set attribute`? well, why? from the stack i guess i'm in metadata, and i'm trying to set an attribute on a RingData instance, but which one? | 00:08 |
| timburke_ | maybe i just need to spend more time with it :-) | 00:10 |
| timburke_ | given the dev_id_bytes variability, do we want to get https://review.opendev.org/c/openstack/swift/+/808532 in to make testing the v2 format a little easier? | 00:30 |
| mattoliver | yeah. the RingSerialization classes and the register (like storage polices), is really cool at defining new things and not have to modify ring serialisation code (look at the last patch adding the dispersion graph). But there could be some better error logging and better divisions, for example to pull off ring like structures we need the dev_id_bytes and the _part_shift, which are both stored in the metadata. So we now have a | 01:24 |
| mattoliver | dependency to deserialize metadata before any ringlike data... which is a little messy. | 01:24 |
| mattoliver | tempted to store them with the ringdata, but we also need it in metadata for when we pull metadata_only. | 01:26 |
| mattoliver | If we combine the ring and the builder. We could combine the metadata (or keep them seperate so a RingData.load() only get's ring data) | 01:27 |
| mattoliver | I'd also expect we only really need one save, RingBuilder.save() and 2 loads RingData.load() (used in the ring to only get the ring components) and RingBuilder.load() to get everything. But arguably we could instead just have 1 load (somewhere, maybe ring.utils) that takes in a list of indexes to load or load everything. | 01:28 |
| mattoliver | I guess we could add some dependency system to Serialization classes. So ringdata pulls in metadata. Or we further seperate ringlike data metadata and keep it seperate and can be a dependency of metadata and ringlike structures (replica2part2dev and past2part2*) | 01:31 |
| opendevreview | Merged openstack/swift master: Ignore py36 deprecation warnings https://review.opendev.org/c/openstack/swift/+/839828 | 04:00 |
| opendevreview | Tim Burke proposed openstack/swift master: Log signal handling at notice https://review.opendev.org/c/openstack/swift/+/789036 | 15:49 |
| timburke__ | strangest gate fixer i've seen: https://review.opendev.org/c/openstack/swift/+/839828 -- cut the execution time for probe tests roughly *in half*, back to where they were a few days ago | 15:53 |
| opendevreview | Tim Burke proposed openstack/swift master: Log signal handling at notice https://review.opendev.org/c/openstack/swift/+/789036 | 16:27 |
| opendevreview | Alistair Coles proposed openstack/swift master: WIP: Round s3api listing LastModified to integer resolution https://review.opendev.org/c/openstack/swift/+/839975 | 17:06 |
| opendevreview | Merged openstack/swift master: Drop arm64 probe test job https://review.opendev.org/c/openstack/swift/+/839785 | 19:02 |
| opendevreview | Merged openstack/swift master: Clarify that rsync_io_timeout is also used for contimeout https://review.opendev.org/c/openstack/swift/+/839797 | 22:29 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!