[Update (2019/10/19 21:02:00 UTC): Both sodium and fluorine have rebooted. Next up will be beryllium rebooting 1d12h from now. That leaves rebooting of helium 18h later and 3h after that will have boron being rebooted. Again, any impact visible to the community should be minimal. See TMB's note, below. --martyb]
We have just learned that Linode, the provider of SoylentNews' server infrastructure, is planning a number of reboots.
[TMB Note]: This shouldn't mean any downtime for anything user-facing except IRC. There will be a few minutes where the comment counts won't update on the front page but those aren't realtime anyway and a few minutes where subscription updates will be delayed until the server that processes them comes back up.
Recently, we identified a commit to the upstream Linux kernel[1] as the cause of an increase in emergency maintenance on our platform. After implementing, testing, deploying, and gaining confidence in a fix, we are now ready to roll this update out to the remainder of our fleet. We're confident this will resolve the bug and ultimately lessen the amount of unplanned maintenance for your Linodes as a result of this specific issue.
To complete this, we will be performing maintenance on a subset of Linode's host machines. This maintenance will update the underlying infrastructure that Linodes reside on and will not affect the data stored within them.
If you are on an affected host, your maintenance window will be communicated to you via a Support ticket within the next few days. You can prepare your Linode for this maintenance by following our Reboot Survival Guide[2].
During the actual maintenance window, your Linode will be cleanly shut down and will be unavailable while we perform the updates. A two-hour window is allocated, however the actual downtime should be much less. After the maintenance has concluded, each Linode will be returned to its last state (running or powered off).
This status page will be updated once maintenance is complete.
[1] https://lkml.org/lkml/2019/1/8/905
[2] https://linode.com/docs/uptime/reboot-survival-guide/
The first server reboot is currently scheduled for Friday, 2019-10-18 at 05:00:00 UTC.
Read on after the fold for more details on the scheduled maintenance dates and times.
Note: All dates and times are in UTC:
Affected systems:
lithium No Maintenance Required Linode 4GB magnesium No Maintenance Required Linode 2GB (pending upgrade) sodium 2019-10-18 05:00 AM Linode 2GB fluorine 2019-10-19 02:00 AM Linode 8GB (pending upgrade) helium 2019-10-22 03:00 AM Linode 8GB hydrogen No Maintenance Required Linode 8GB neon No Maintenance Required Linode 8GB beryllium 2019-10-21 09:00 AM Linode 4GB (pending upgrade) boron 2019-10-22 05:00 AM Linode 4GB (pending upgrade)
(Score: 2) by The Mighty Buzzard on Monday October 07 2019, @02:07PM (4 children)
Yeah, it's a pain for us admins. No downtime you lot should notice unless you're on our IRC server though.
My rights don't end where your fear begins.
(Score: 0) by Anonymous Coward on Tuesday October 08 2019, @04:29AM (3 children)
Is Linode just using KVM?
If yes, does anyone know why they don't just do live migrations?
Shared storage doesn't seem a valid reason for quite a few years now. Ceph's rados block device is reliable, and does not require much more raw disk than they would currently be using. DRBD between pairs of hosts would work, but would require 2x the storage.
(Score: 2) by The Mighty Buzzard on Tuesday October 08 2019, @07:30AM (2 children)
Probably because they're not migrating storage, they're updating kernels on the host OSes. And, frankly, I'd be shopping for new hosting providers if I thought they did live kernel updates on their host boxes.
My rights don't end where your fear begins.
(Score: 0) by Anonymous Coward on Wednesday October 09 2019, @01:43AM (1 child)
KVM allows live migration of VMs (if you have shared storage).
I meant live migration of the VM.
1) migrate VMs off host A to free capacity on some other nodes
2) patch and reboot A
3) migrate from host B to A
4) patch and reboot B
.
.
.
Customer not affected by hosting providers operational issues that customer should not have to know or care about.
(Score: 3, Interesting) by t-3 on Sunday October 20 2019, @01:15AM
Why bother with that when actual downtime is negligible and it's much simpler to just reboot without trying to migrate? If a customer requires 100% uptime, I'm sure they're going to have a plan for that.