*** gmann is now known as gmaan | 02:13 | |
*** gmaan is now known as info | 03:50 | |
*** info is now known as IDENTIFY | 03:52 | |
*** IDENTIFY is now known as gmaan_pto | 03:52 | |
opendevreview | Jay Faulkner proposed openstack/ironic master: Mark SNMP driver unsupported for removal https://review.opendev.org/c/openstack/ironic/+/946843 | 16:23 |
---|---|---|
opendevreview | Amarachi Ordor proposed openstack/bifrost master: docs: updated test guide, improve how-to clarity, add troubleshooting tips https://review.opendev.org/c/openstack/bifrost/+/946953 | 17:25 |
opendevreview | cid proposed openstack/ironic-python-agent master: WIP: Eventlet Removal- WSGI server https://review.opendev.org/c/openstack/ironic-python-agent/+/946091 | 22:59 |
JayF | holy crap cid it works | 23:48 |
JayF | like obviously a few rough edges but that is an ipa job passing without an eventlet wsgi server \o/ | 23:49 |
cid | Indeed :) | 23:52 |
cid | And minimal change, really, fingers crossed | 23:53 |
JayF | yeah | 23:53 |
JayF | you might have earned my attention monday, seeing if I can get it to eventlet zero and bang around the edges some in ironic | 23:53 |
JayF | one thing to test explicitly is where we stream an image from the conductor | 23:54 |
cid | On the Ironic side (?) | 23:54 |
JayF | so in the IPA code | 23:54 |
JayF | the image downloader can either be streaming to disk (getting from net -> write to disk one chunk at a time) | 23:54 |
JayF | or can download it all at once and then write it to disk (net -> ramdisk -> disk) | 23:55 |
JayF | look at the documentation around ... image_download_source? | 23:55 |
JayF | https://docs.openstack.org/ironic/latest/admin/interfaces/deploy.html#streaming-raw-images | 23:56 |
cid | Got it. | 23:56 |
* cid was about to start looking for that docs :D | 23:56 | |
JayF | yeah essentially if you're not using an external http server, and the conductor is being a proxy anyway | 23:57 |
JayF | you enable streaming and you massively reduce the ram needed on the agent machine | 23:57 |
JayF | I am not 100% sure what the exact ratio is, but it was something like you needed 2x the uncompressed size of the kernel+ramdisk + about 250% of the size of the image in question | 23:57 |
JayF | (in ram) | 23:57 |
JayF | but that is from memory from a long enough time ago that it's unreliable lol | 23:58 |
JayF | but it makes sense, that 5 GB image has to go somewhere while it's being downloaded and decompressed, and ram is the only place | 23:58 |
JayF | (IPA will never write to local disk except as a deployment/cleaning action) | 23:58 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!