Stories
Slash Boxes
Comments

SoylentNews is people

Meta
posted by paulej72 on Sunday December 13 2015, @03:57PM   Printer-friendly
from the if-at-first-you-don't-sucseede dept.

Hello fellow Soylentils!

We were informed by Linode (our hosting provider) that they needed to perform Xen updates on all of our servers. This forces a reboot of our virtual servers which may cause the site (and other services) to be temporarily unavailable.

Here is the two-day reboot schedule along with what runs on each server:

Status Day Date Time Server Affects
Done Mon 2015-12-14 0300 UTC lithium Development Server, slashd, Varnish, MySQL, Apache
Done Mon 2015-12-14 0400 UTC helium Production Back End, MySQL NDB, DNS, Hesoid, Kerberos
Done Mon 2015-12-14 0400 UTC beryllium IRC, MySQL, Postfix, Mailman, Yourls
Done Mon 2015-12-14 0400 UTC sodium Primary Load Balancer
Done Mon 2015-12-14 0400 UTC magnesium Backup Load Balancer
Done Mon 2015-12-14 0700 UTC boron DNS, Hesoid, Kerberos, Staff Slash
Done Tue 2015-12-15 0300 UTC fluorine Production Front End, slashd, Varnish, MySQL, Apache, ipnd
Done Tue 2015-12-15 0300 UTC neon Production Back End, MySQL NDB cluster
Done Tue 2015-12-15 0400 UTC hydrogen Production Front End, Varnish, MySQL, Apache, Sphinx

As we found some errors with our MySql setup during the last Xen updates, The Main Site will be down starting at 2015-12-14 at 0300 UTC while we reconfigure MySQL prior to the shutdown of the main MySQL server helium. The site should be up by 0500 UTC.

We apologize in advance for any inconvenience and appreciate your understanding as we try and get things up and running following each reboot.

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by Runaway1956 on Sunday December 13 2015, @04:02PM

    by Runaway1956 (2926) Subscriber Badge on Sunday December 13 2015, @04:02PM (#275786) Homepage Journal
    • (Score: 2) by Knowledge Troll on Sunday December 13 2015, @04:48PM

      by Knowledge Troll (5948) on Sunday December 13 2015, @04:48PM (#275798) Homepage Journal

      I looked a Hesiod the last time I needed central auth but stuck with LDAP. I'm glad to see it in use somewhere and I'm wondering how it has worked out for you. Looked like a great idea if it would do what was needed and DNSSEC was enabled.

      Having Kerberos on the naming server though seems weird. The kind of weird where a bug in BIND (or dns server of choice) followed up with a bug in the OS Kernel causes the entire auth system to be untrustable.

      I am really curious though how this architecture has worked out.

    • (Score: 0) by Anonymous Coward on Sunday December 13 2015, @11:25PM

      by Anonymous Coward on Sunday December 13 2015, @11:25PM (#275892)

      No, they just use their NGO and banks and various organisations to undermine western societies.

  • (Score: 0) by Anonymous Coward on Sunday December 13 2015, @04:58PM

    by Anonymous Coward on Sunday December 13 2015, @04:58PM (#275799)

    Got a similar email from Linode. They wrote that I could avoid the reboot by switching to KVM. Thoughts? Does KVM need as many reboots as Xen?

    • (Score: 2) by isostatic on Sunday December 13 2015, @07:33PM

      by isostatic (365) on Sunday December 13 2015, @07:33PM (#275833) Journal

      Fewer security problems perhaps?

      A modern resiliant installation spreads out across multiple nodes on multiple continents and can survive many of them vanishing without notice. It can even self heal as the number of nodes for a given task reduces, so new ones are created.

      The stickler tends to be DNS over multiple networks, you could get an outage of a few minutes if you have say a ttl of 300 and you lose the IP address DNS points to.

    • (Score: 0) by Anonymous Coward on Sunday December 13 2015, @08:47PM

      by Anonymous Coward on Sunday December 13 2015, @08:47PM (#275844)

      Probably. I'll be getting the linode email at work on Monday. I'm closing down the Linode VM's and moving to hardware virtualization with another provider precisely so I don't have to use xen kernels - even if performance is a little worse than under xen.

    • (Score: 2) by paulej72 on Monday December 14 2015, @02:13AM

      by paulej72 (58) on Monday December 14 2015, @02:13AM (#275970) Journal

      We need to test if our setup is KVM compatible. As they did not give us much time to prepare, I thought it best to hold off on that migration.

      --
      Team Leader for SN Development
      • (Score: 1, Informative) by Anonymous Coward on Monday December 14 2015, @10:13AM

        by Anonymous Coward on Monday December 14 2015, @10:13AM (#276045)

        I converted our 5 remaining Xen Linode VPSs to KVM this weekend. No issues whatsoever. Plus, the downtime was much shorter than the Xen maintenance window.

        The response time on KVM seems to be improved as well. It's certainly something SN should consider.

  • (Score: -1, Offtopic) by Anonymous Coward on Sunday December 13 2015, @05:03PM

    by Anonymous Coward on Sunday December 13 2015, @05:03PM (#275800)

    I LOVE TO RUB MY BALLS ON MY GRANDMA'S FOREHEAD!

    • (Score: -1, Troll) by Anonymous Coward on Sunday December 13 2015, @05:12PM

      by Anonymous Coward on Sunday December 13 2015, @05:12PM (#275801)

      RIP in pieces Grandma
            1927-2016
      died of testicle overdose

  • (Score: 2) by boltronics on Monday December 14 2015, @02:07AM

    by boltronics (580) on Monday December 14 2015, @02:07AM (#275968) Homepage Journal

    I don't know about Linode, but AWS (and surely other provides) gives you the option to launch instances onto Xen hosts that already have the update applied. That means you can migrate your instances to new hosts in advance of the reboot, and avoid downtime - assuming you are using a configuration management system to seamlessly deploy new instances (which is a generally a really good idea). That's how we manage our websites, the bulk of which is spread across 21 instances.

    --
    It's GNU/Linux dammit!
    • (Score: 2) by paulej72 on Monday December 14 2015, @02:15AM

      by paulej72 (58) on Monday December 14 2015, @02:15AM (#275973) Journal

      Our setup is a bit too small to have configuration management. It would be more work than what it would get us.

      --
      Team Leader for SN Development
    • (Score: 2) by zeigerpuppy on Thursday December 17 2015, @04:31AM

      by zeigerpuppy (1298) on Thursday December 17 2015, @04:31AM (#277516)

      My setup is a bit small for this too,
      But am interested in what you suggest for configuration management.
      I assume that the primary issues would transitioning DNS redirections, http proxy configs and ensuring database continuity.
      On my Xen server I currently manage things by having a management VM that acts as a salt master but wondering what tools you recommend for more complex setups?

      • (Score: 3, Informative) by boltronics on Thursday December 17 2015, @06:47AM

        by boltronics (580) on Thursday December 17 2015, @06:47AM (#277566) Homepage Journal

        We're using Salt Stack as well.

        I create clean AMI base images using Debian Image Builder [github.com] (currently without systemd), and share those AMIs from our production account to our separate AWS staging accounts for testing. All instances everywhere are then launched from the same AMI which we fully trust since we built it.

        We have salt-cloud configured to provision instances as required, as we do not (as of yet) auto-scale. We do have a Orchestrate setup in place, but haven't had much of a need to use it - instead preferring to replace just part of our infrastructure at a time (such as the backend app nodes and memcached servers, and that's just a matter of spinning up new replacement instances. In the case of app nodes, the latest builds are deployed from git directly or (depending on the application) from our CI server which builds debs and pushes them to an apt repo when all tests pass. When we're satisfied all the new app nodes are happy, we can nuke the old ones from the Salt Mine, re-run highstate on the front-end proxies (which regenerates the configs), so the load is now distributed exclusively to the new servers. We can then just terminate the old instances at our leisure.

        Front end proxies are migrated by just firing up new ones and jumping across the AWS Elastic IP. We use multi-AZ RDS instances for databases, so typically we can just fail-over to another zone and not have to do much there.

        We don't reference hosts by DNS internally in most cases, opting to use AWS tags and Salt grains instead (which are all auto-assigned via salt-cloud). IPs of instances are stored in the Mine and easily queried. We mainly just use DNS for public-facing addresses, and those remain static due to our use of Elastic IPs when required.

        --
        It's GNU/Linux dammit!
        • (Score: 2) by zeigerpuppy on Thursday December 17 2015, @09:31PM

          by zeigerpuppy (1298) on Thursday December 17 2015, @09:31PM (#277912)

          Thanks for the description. Useful for when our service expands.
          At the moment in loving the combination of Xen and ZFS on our Debian servers.
          I can clone a DomU really quickly (their root partitions are zvols) and it's easy snapshotting them.
          However we are using internal DNS which I'm
          Not sure if it's a strength or weakness yet!

          • (Score: 2) by boltronics on Friday December 18 2015, @12:48AM

            by boltronics (580) on Friday December 18 2015, @12:48AM (#278026) Homepage Journal

            Makes sense.

            For instances I host for personal use (on my home server running my blog, e-mail, etc.), or for infrastructure hosted on my workplace premises, I use Xen instances too. I use LVM2 though instead of ZFS, which works nicely with xen-tools on Wheezy and I feel is better supported for now. But everything in those environment is pretty much static (and I can always easily schedule downtime after hours) so I do use a DNS server in those situations for instance addressing.

            To be fair, I could use DNS for IaaS-hosted instances too, but it seems redundant and adds unnecessary complexity when there are easier ways.

            --
            It's GNU/Linux dammit!