| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/zuul] 920600: Fix change cache upgrade https://review.opendev.org/c/zuul/zuul/+/920600 | 01:03 | |
| @anoop-jose:matrix.org | Guys, how far zuul support plugins? | 03:28 |
|---|---|---|
| @anoop-jose:matrix.org | Is there any guidelines or references you can share about plugin development of zuul ? | 03:29 |
| @fungicide:matrix.org | @anoop-jose:matrix.org: what do you mean by "plugins"? | 13:05 |
| @fungicide:matrix.org | zuul calls ansible, so can perform pretty much any task ansible is capable of | 13:06 |
| @fungicide:matrix.org | ansible itself has a versatile language for composing basic functionality and comes with modules for most common sorts of things, but you can also extend its language by writing your own modules | 13:08 |
| @f2ked:matrix.org | I have list of 100s of tests that I'd like to distribute over a set of nodes. The segmentation (eg. `[[t0, t10, t9,...], [t3, t4, t20,...],...]`) is created dynamically (via a script) that tries to balance (uniform running time) them based on prior performance results. | 13:22 |
| What is the "Zuul way" arranging for each segment to run on a node. | ||
| Can `Jobs` be dynamically created? (how we do it in jenkins today) | ||
| @jim:acmegating.com | f2ked: no dynamic nodes or jobs, but you can try making a single multi-node job and distributing the tests to the nodes of that job. | 13:24 |
| @jim:acmegating.com | * f2ked: no dynamic nodes or jobs, but you can try making a single multi-node job and distributing the tests to the nodes of that job (using ansible to cause the execution on the multiple nodes) | 13:30 |
| @f2ked:matrix.org | corvus: That makes sense, thanks! | 13:33 |
| @clarkb:matrix.org | You can also theoretically use tools like stestr or bazel etc to do the balancing for you | 14:55 |
| @clarkb:matrix.org | so basically just configure a tool like that to talk to a number of nodes in a job and then as long as you can seed it with some timing info it should balance? | 14:56 |
| @jim:acmegating.com | yep (you would use zuul's ansible to configure the nodes to receive testr/bazel instructions) | 15:06 |
| @sylvass:albinvass.se | Clark: ++, which is generally why I like to separate the ci/cd system and the build system. They have different goals. | 15:17 |
| @jim:acmegating.com | (or of course use a bazel remote execution thing, or distcc, or...) | 15:18 |
| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/zuul] 920690: Pre-filter gerrit events based on triggers https://review.opendev.org/c/zuul/zuul/+/920690 | 17:10 | |
| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/nodepool] 920691: Use cached node list in createMinReady https://review.opendev.org/c/zuul/nodepool/+/920691 | 18:16 | |
| -@gerrit:opendev.org- James E. Blair https://matrix.to/#/@jim:acmegating.com proposed: [zuul/zuul] 920693: Add buildset events to buildset summary page https://review.opendev.org/c/zuul/zuul/+/920693 | 18:58 | |
| @fungicide:matrix.org | on that note, it would be a cool weekend project to add roles for traditional matrix processing/message passing frameworks and have zuul dynamically create a beowulf cluster to run embarrassingly parallel jobs | 19:13 |
| @fungicide:matrix.org | supporting on-demand pvm or mpi(ch) based workloads | 19:13 |
| @fungicide:matrix.org | though i suppose hpc type workloads like that are better served by persistent bare metal clusters | 19:15 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!