Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2025-03-18
16:32:14 elodilles +2 from me too
16:32:15 sean-k-mooney and dicuss it ther
16:32:32 sean-k-mooney ishanwar[m]: we will loop back ot you next
16:32:38 bauzas cool cool, then I'll propose myself :)
16:32:51 bauzas that was short and quick :)
16:32:56 ishanwar[m] sean-k-mooney: sure
16:33:30 Uggla so I think we can discuss ishanwar[m] topic.
16:33:48 Uggla ishanwar[m], please go ahead.
16:34:29 ishanwar[m] I wanted to have a discussion on https://bugs.launchpad.net/nova/+bug/2027156
16:34:35 dansmith that doesn't look like low-hanging fruit to me :)
16:34:47 dansmith is there a review up?
16:35:05 ishanwar[m] no I am still working on it
16:35:38 dansmith okay, if the quota api is not something that we can guarantee is available, then we have to have all the code we have today in nova, plus new code to use the api if present, right?
16:35:40 ishanwar[m] wanted to discuss if I understood it correctly,... (full message at <https://matrix.org/oftc/media/v1/media/download/AZZiNbMs1GZbZLUOwiWvG2I_Dg7_qJ2sHecbqrafIK6K_9GcmEgBnqAHwcu-vfVKIV_6lx6PmdvT59UIS_UEVuBCeV8jVjWgAG1hdHJpeC5vcmcvekZxaUt3b3hkYnpmanJ2dkZqQ2lESFJY>)
16:36:30 ishanwar[m] as the candidate for change but i am not sure, if we are counting in ports in that function.
16:36:30 ishanwar[m] the bug mentions https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L2654
16:36:39 ishanwar[m] am i right ?
16:37:15 sean-k-mooney so part of the problem is the orginal bug report does not use permalinks
16:37:28 sean-k-mooney so they dont nessiarly link to the correct code any more
16:37:28 dansmith ishanwar[m]: that bug was filed by a neutron person so I suspect we won't be able to answer your question authoritatively without really digging in
16:38:08 sean-k-mooney ishanwar[m]: you are correct that validate_networks does use the quota api already https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L2714
16:38:50 sean-k-mooney i guess the real quetion is do we use that in all places we do quota checks
16:39:05 sean-k-mooney this was orginaly reported downstream for trian
16:39:13 sean-k-mooney so its very possibel that htis was fixed already in master
16:39:25 ishanwar[m] Yes correct, but it's still listing all ports to count them, rather than using show_quota_details, right ?
16:40:07 sean-k-mooney we do a port list in get_requested_resource_for_instance
16:40:43 sean-k-mooney but that not listing all ports for the porject
16:40:49 ishanwar[m] sean-k-mooney: not sure if this is fixed elsewhere, I was only able to find one usage of len(ports)
16:40:52 sean-k-mooney that is geting port for the current instnace wihch is fine
16:41:03 ishanwar[m] sean-k-mooney: yes
16:41:16 dansmith we fetch the quota, but we do our own usage calculation it looks like
16:41:31 sean-k-mooney oh its this second call https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L2721
16:41:42 dansmith we don't ask if the quota is exceeded, we ask what the quota is and then sum the current+new and see if it's over
16:41:45 sean-k-mooney that ^ is the line that shoudl be updated
16:41:49 ishanwar[m] dansmith: yes in https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L2714
16:41:49 ishanwar[m] we are still counting all the ports
16:42:06 sean-k-mooney ishanwar[m]: so the bug is this
16:42:23 sean-k-mooney there is another extention that gives use both the quota and the uses cont in one call
16:42:34 ishanwar[m] yes correct, i think we can use show_quota_details here right ?
16:42:42 ishanwar[m] sean-k-mooney: which one ?
16:42:48 sean-k-mooney so instead of getting all the ports for the curent project just to count them we can use show_quota_details
16:43:02 sean-k-mooney ishanwar[m]: so the performace bug is still there
16:43:04 dansmith ...if available
16:43:07 sean-k-mooney on this line https://github.com/openstack/nova/blob/6042300453411133652dd8d3b789f5057a084075/nova/network/neutron.py#L2721
16:43:24 sean-k-mooney but ya as dan impleid we have to fall back if the extension is not there
16:43:45 dansmith it's also racy either way, which is unfortunate
16:43:55 ishanwar[m] so we can use a try catch mech to check if the extension is available or not
16:43:55 ishanwar[m] sean-k-mooney: correct I agree
16:44:06 ishanwar[m] dansmith: oh, how is that ?
16:44:27 dansmith ishanwar[m]: if we calculate the quota, allow to proceed and other things in parallel use more quota, we'll be wrong and run out of quota while actually creating the ports
16:45:06 dansmith since we already need to handle that, why check quota at all? Just try to create the ports and roll back if you run out.. I guess maybe to provide a quick answer via API, but it will never be guaranteed
16:45:25 sean-k-mooney well we do prot creation on the comptues
16:45:36 sean-k-mooney so we do the api check to fail fast if we are already at quota
16:45:40 dansmith this can say "definitely not enough quota" but it can't say "definitely enough"
16:45:50 dansmith right
16:46:30 ishanwar[m] Not sure I understood 😓
16:46:45 dansmith there's a long window between checking quota and consuming it
16:46:45 sean-k-mooney ishanwar[m]: so to answer your question, the bug is still there adn we can still optimize this call to neturon
16:46:59 sean-k-mooney but we need to fallback if the extion is not there to the current logic.
16:47:58 ishanwar[m] I have another question, i am running the openstack nova dev env on devstack. is there a way how I can test my changes for this bug, sorry I am new to this ?
16:48:53 dansmith with devstack yes, make your changes and test manually, but this will also need unit/functional tests in the patch
16:49:37 sean-k-mooney ishanwar[m]: to do ^ you just modify the code and do "systemctl restart devstack@n-*"
16:49:50 sean-k-mooney i.e. modify the code in /opt/stack/nova
16:50:53 Uggla ishanwar[m], also finding something similar to the bug in gerrit can be a good example to know what the patch should have in terms of unit/fn tests.
16:50:53 sean-k-mooney i generally do it slightly idffently by have a copy in another location and pip install that into the devstack virutal env but you can basiclly just modify the code and restart the affected services. in thsi case devtack@n-api is the nova api
16:51:33 ishanwar[m] this restarts all nova services right ? Shouldn't we check just one, in general what process do you follow ? Do you add breakpoints etc for testing things out ?
16:51:40 sean-k-mooney you can look at the function above https://github.com/openstack/nova/blob/6042300453411133652dd8d3b789f5057a084075/nova/network/neutron.py#L2672-L2693
16:51:48 sean-k-mooney it does an if/else on if the extenion is supproted
16:52:07 sean-k-mooney you can look at howe has_extended_resource_request_extension is implemnted to test for the quta details one
16:52:10 ishanwar[m] sean-k-mooney: yes understood, makes sense .
16:52:41 sean-k-mooney ishanwar[m]: so you cant really run nova under a debugger
16:52:46 sean-k-mooney its posssible but hard
16:52:52 sean-k-mooney so generally we add debug logs
16:52:53 ishanwar[m] sean-k-mooney: yeah I can implement ifelse for this
16:53:13 sean-k-mooney that useful for production as well as in production we will generally only have logs aviable
16:53:43 ishanwar[m] sean-k-mooney: okay I see.
16:54:44 ishanwar[m] sean-k-mooney: hmm i see. So from what I understood, add logs, for testing, restart the n-api service. Use openstack command line for testing, correct ?
16:55:06 sean-k-mooney initally but then you will need to add unit test for each branch
16:55:20 sean-k-mooney so mocking the has_extenion_* function for true and false
16:55:37 ishanwar[m] Can you share an example ?
16:55:40 sean-k-mooney and assert list_port or quota_detials is called approreatly
16:56:00 sean-k-mooney yep we can proably do that perhaps after the meeting
16:56:05 sean-k-mooney incase there are any other topics
16:56:35 ishanwar[m] sure, thanks sean-k-mooney dansmith Uggla
16:56:35 ishanwar[m] i think we can close this discussion
16:56:40 ishanwar[m] back to you Uggla
16:56:49 Uggla thx @ishanwar[m]
16:57:15 Uggla any other topic ?
16:58:25 Uggla looks we can close the meeting
16:58:34 Uggla 3
16:58:51 bauzas shoot :)
16:58:53 Uggla 2
16:58:55 dansmith blastoff
16:59:05 Uggla thanks all
16:59:10 opendevmeet Log: https://meetings.opendev.org/meetings/nova/2025/nova.2025-03-18-16.01.log.html
16:59:10 opendevmeet Minutes (text): https://meetings.opendev.org/meetings/nova/2025/nova.2025-03-18-16.01.txt
16:59:10 opendevmeet Minutes: https://meetings.opendev.org/meetings/nova/2025/nova.2025-03-18-16.01.html
16:59:10 opendevmeet Meeting ended Tue Mar 18 16:59:10 2025 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)
16:59:10 Uggla #endmeeting
16:59:17 bauzas heh, that worked well :)

Earlier   Later