public inbox for pve-devel@lists.proxmox.com
 help / color / mirror / Atom feed
From: "Daniel Kral" <d.kral@proxmox.com>
To: "Proxmox VE development discussion" <pve-devel@lists.proxmox.com>
Cc: "pve-devel" <pve-devel-bounces@lists.proxmox.com>
Subject: Re: [pve-devel] [PATCH ha-manager 02/18] manager: retranslate rules if nodes are added or removed
Date: Wed, 27 Aug 2025 16:55:07 +0200	[thread overview]
Message-ID: <DCDAP30FN14M.7U6R0GNY2MVQ@proxmox.com> (raw)
In-Reply-To: <20250821143705.256562-3-d.kral@proxmox.com>

On Thu Aug 21, 2025 at 4:35 PM CEST, Daniel Kral wrote:
> Some rule checks depend on the list of cluster nodes, e.g., to check
> whether a negative resource affinity rule doesn't specify more HA resources than cluster nodes.
>
> The HA Manager retranslate rules only in certain conditions to reduce
> unnecessary computations, but lacks a check whether cluster nodes have
> been added or removed, which is different from what users are reported
> through the rules API endpoints and web interface.
>
> Fixes: 6c4c0458 ("rules: add haenv node list to the rules' canonicalization stage")
> Signed-off-by: Daniel Kral <d.kral@proxmox.com>
> ---
>  src/PVE/HA/Manager.pm    |  2 ++
>  src/PVE/HA/NodeStatus.pm | 14 ++++++++++++++
>  2 files changed, 16 insertions(+)

As @Michael and I briefly taked about this off-list, the nodelist
shouldn't cange too much in production (i.e. the PVE2 environment), but
this check makes the HA rules retranslation more correct as the checks
are dependent on $nodes.

AFAICT the main reasons the nodelist changes in production is that a
node joins or leaves, where PVE::HA::Env::PVE2::get_node_info($self)
gets the nodelist from PVE::Cluster::get_members().

Even though the pve-ha-crm systemd unit has an ordering dependency on
pve-cluster, pvedaemon, ..., which are restarted on node join, these are
only restarted on the newly added node AFAICS when calling
PVE::Cluster::Setup::join, so the HA Manager isn't updated with the new
nodelist in that case, therefore the added condition in this patch is
required.

I noticed this when implementing the plugin_compile for the node
affinity rules, which heavily depend on the $nodes. Without this
additional condition, at least 'test-crs-static2' will fail as it
doesn't power on all nodes at once, but only powers node4 some time
later. The nodelist won't be updated and therefore the HA node affinity
rule in that test cases won't get retranslated too, which will change
the behavior of the node (re)assignment.

The check helper could definitely been implemented nicer, but I didn't
want to overcomplicate things here.


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel


  parent reply	other threads:[~2025-08-27 14:55 UTC|newest]

Thread overview: 34+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-08-21 14:35 [pve-devel] [PATCH ha-manager 00/18] HA rules fixes + cleanup + performance improvements Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 01/18] config: do not add ignored resources to dependent resources Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 02/18] manager: retranslate rules if nodes are added or removed Daniel Kral
2025-08-27 13:41   ` Michael Köppl
2025-08-27 14:33     ` Daniel Kral
2025-08-27 14:55   ` Daniel Kral [this message]
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 03/18] rules: factor out disjoint rules' resource set helper Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 04/18] rules: resource affinity: inter-consistency check with merged positive rules Daniel Kral
2025-08-29 12:43   ` Michael Köppl
2025-08-29 13:28     ` Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 05/18] rules: add merged positive resource affinity info in global checks Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 06/18] rules: make rules sorting optional in foreach_rule helper Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 07/18] rename rule's canonicalize stage to transform stage Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 08/18] rules: make plugins register transformers instead of plugin_transform Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 09/18] rules: node affinity: decouple get_node_affinity helper from Usage class Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 10/18] compile ha rules to a more compact representation Daniel Kral
2025-08-27 15:19   ` Daniel Kral
2025-08-29 12:43   ` Michael Köppl
2025-08-29 13:42     ` Daniel Kral
2025-09-02  7:32       ` Michael Köppl
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 11/18] test: rules: use to_json instead of Data::Dumper for config output Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 12/18] test: rules: add compiled config output to rules config test cases Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 13/18] rules: node affinity: define node priority outside hash access Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 14/18] move minimum version check helper to ha tools Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 15/18] manager: move group migration cooldown variable into helper Daniel Kral
2025-08-29 12:43   ` Michael Köppl
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 16/18] api: status: sync active service counting with lrm's helper Daniel Kral
2025-09-02  8:10   ` Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 17/18] manager: group migration: " Daniel Kral
2025-08-21 14:35 ` [pve-devel] [PATCH ha-manager 18/18] factor out counting of active services into helper Daniel Kral
2025-08-29 12:44   ` Michael Köppl
2025-08-29 13:36     ` Daniel Kral
2025-08-29 12:44 ` [pve-devel] [PATCH ha-manager 00/18] HA rules fixes + cleanup + performance improvements Michael Köppl
2025-08-29 13:52   ` Daniel Kral

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=DCDAP30FN14M.7U6R0GNY2MVQ@proxmox.com \
    --to=d.kral@proxmox.com \
    --cc=pve-devel-bounces@lists.proxmox.com \
    --cc=pve-devel@lists.proxmox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
Service provided by Proxmox Server Solutions GmbH | Privacy | Legal