From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from gate001.proxmox.com (gate001.proxmox.com [IPv6:2a0f:8001:1:32::40]) by lore.proxmox.com (Postfix) with ESMTPS id EC58E1FF0BC for ; Mon, 16 Nov 2026 13:40:20 +0100 (CET) Received: from gate001.proxmox.com (localhost.localdomain [127.0.0.1]) by gate001.proxmox.com (Proxmox) with ESMTP id CBC8721718; Mon, 16 Nov 2026 13:40:16 +0100 (CET) Mime-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=UTF-8 Date: Mon, 16 Nov 2026 13:40:12 +0100 Message-Id: From: "Daniel Kral" To: =?utf-8?q?Michael_K=C3=B6ppl?= , Subject: Re: [PATCH ha-manager v2 12/16] lrm: update the service status once per work iteration X-Mailer: aerc 0.22.0-45-gc30c872fde7b-dirty References: <20261112163113.294302-3-d.kral@proxmox.com> <20261112163113.294302-15-d.kral@proxmox.com> In-Reply-To: X-Bm-Milter-Handled: 55990f41-d878-4baa-be0a-ee34c49e34d2 X-Bm-Transport-Timestamp: 1794832812450 X-SPAM-LEVEL: Spam detection results: 0 AWL 0.565 Adjusted score from AWL reputation of From: address DMARC_MISSING 0.1 Missing DMARC policy KAM_DMARC_STATUS 0.01 Test Rule for DKIM or SPF Failure with Strict Alignment (newer systems) RCVD_IN_DNSWL_MED -2.3 Sender listed at https://www.dnswl.org/, medium trust SPF_HELO_NONE 0.001 SPF: HELO does not publish an SPF Record SPF_PASS -0.001 SPF: sender matches SPF record Message-ID-Hash: 4X5EUBA4FNUQUYRNRRGRNEUBJUZWRDWG X-Message-ID-Hash: 4X5EUBA4FNUQUYRNRRGRNEUBJUZWRDWG X-MailFrom: d.kral@proxmox.com X-Mailman-Rule-Misses: dmarc-mitigation; no-senders; approved; loop; banned-address; emergency; member-moderation; nonmember-moderation; administrivia; implicit-dest; max-recipients; max-size; news-moderation; no-subject; digests; suspicious-header X-Mailman-Version: 3.3.10 Precedence: list List-Id: Proxmox VE development discussion List-Help: List-Owner: List-Post: List-Subscribe: List-Unsubscribe: On Mon Nov 16, 2026 at 11:32 AM CET, Michael K=C3=B6ppl wrote: > On Thu Nov 12, 2026 at 5:31 PM CET, Daniel Kral wrote: > > [snip] > >> diff --git a/src/PVE/HA/LRM.pm b/src/PVE/HA/LRM.pm >> index 7ae239ec..a68ab008 100644 >> --- a/src/PVE/HA/LRM.pm >> +++ b/src/PVE/HA/LRM.pm >> @@ -385,7 +385,7 @@ sub work { >> my $status =3D $self->get_local_status(); >> my $state =3D $status->{state}; >> >> - $self->update_lrm_status(); >> + my $status_update_successful =3D $self->update_lrm_status(); >> >> my $fence_request =3D $self->is_fence_requested(); >> >> @@ -516,7 +516,7 @@ sub work { >> >> # if we could not get the current service status there's no= point >> # in doing anything, try again next round. >> - return if !$self->update_lrm_status(); >> + return if !$status_update_successful; >> >> if ($self->{shutdown_request}) { >> >> @@ -615,7 +615,7 @@ sub work { >> } elsif ($state eq 'maintenance') { >> >> my $startime =3D $haenv->get_time(); >> - return if !$self->update_lrm_status(); >> + return if !$status_update_successful; > > I noticed that in the 'active' branch, this happens inside an eval, > resulting in a `return 1` from the `work()` function, whereas here it > would `return undef`. Is there a reason for this? Because just from > reading through the `work()` function and where it's used (haven't > tried this yet) it seems like this would stop the LRM entirely? On the > master this would happen if the second call of `update_lrm_status()` > failed for some reason. Here, I think the following could happen: > > # pmxcfs is temporarily unavailable, so this is undef > my $status_update_successful =3D $self->update_lrm_status(); > > # pmxcfs is up again within 5s (the max wait time for getting the > # lock) > > # this works now > get_protected_ha_agent_lock(); > > ... > > # in maintenance branch > return if !$status_update_successful; # returns due to stale status > > This doesn't introduce the problem, but widens then window for it, no? > I'll try to test this as well. I don't think this is very likely to > occur (considering pmxcfs has to be unwritable and then writable again > within a ~5s window), but it could also easily be avoided, unless I'm > missing something about why there's no eval in for the maintenance > block? Yes, the automated review contained this as well... This would shutdown the LRM and seems like to originate from the active branch above. I haven't fixed it here yet, because another patch series that focuses on the LRM maintenance mode (not sent yet though) touches the same part as well and seemed like it would fit better there. Plus, putting it in a separate patch doesn't put any pressure on which patch series should be applied first. However, I can fix it here already. On another note, the case where the first update_lrm_status() failed, but get_protected_ha_agent_lock() and the second update_lrm_status() in active/maintenance branch succeed is probably a bit odd, because the state transition is based on the old data from the previous LRM round while the actual work is being done in the new data. I don't see where this could make entirely wrong decisions (the obvious would probably be 'self-healed' in the next round then), but might be notable in the patch message itself.