August 2018 Power Outage Plan
There is a planned power outage in MC from Tuesday, August 21 to Thursday, August 30.
There is also a one-day outage in DC, which will complicate keeping services up during the entire outage.
All services in MC for Aug. 21-30, and services in DC for two days during that window.
Services in PHY are not affected (This is redundant DNS and Authentication services. There are no other services (general-use or otherwise) in PHY.). It's also on a different network then MC and DC.
Before Sunday, August 19
- Complete plan for outage
- Send notifications (and reminders) to csc-general
- Take backups of LDAP and Kerberos, and download offsite
- Take backups of system passwords, and download offsite
- Take backups of important containers/machines (whole things or just config): auth1, mail, caffeine
Sunday, August 19
- Copy the CSC website to caffeine-dr
- Shutdown general-use computing services
- Shutdown csclub.cloud components (they won't really work since not everything is redundant yet)
- Transfer computing services to redundant / temporary systems
- Revoke access to home directories on aspartame to all machines
Sometime during the outage window
- Shutdown DC systems before the building outage
After the outage
- Being restoring normal services
Our network is announced from both MC and DC. No impact to networking is expected when MC goes offline.
DHCP is hosted in MC (on caffeine). This is not strictly required as our servers use static IPs, but we can move it to DC so it's available.
CSCF will provided some generator power for mirror in MC.
CSCF is also setting up a second node in DC.
A copy of the CSC website will be hosted on caffeine-dr. All pages not found on the local machine (including member and club sites) will return a 503 Service Unavailable error page.
Sample status page: https://www-dr.csclub.uwaterloo.ca/test
The following IP addresses should be added to caffeine-dr during the outage to serve the error page for other CSC services:
- caffeine: 18.104.22.168 / 2620:101:f000:4901:c5c::caff:e12e
- git: 22.214.171.124 / 2620:101:f000:4901:c5c:3eb::49
- wiki: 126.96.36.199 / 2620:101:f000:4901:c5c:3eb::44
- munin: 188.8.131.52 / 2620:101:f000:4901:c5c::51
- prometheus: 184.108.40.206 / 2620:101:f000:4901:c5c::15
Since the outage is for a week, we need to maintain email services during the outage. An initial plan by ztseguin and jxpryde:
- rsync users' .forward, .procmailrc and .maildir to a local directory on mail, allowing mail to continue as expected
However, this requires:
- Users not reference any scripts, programs, etc. in their procmailrc file that reference things in their home directory
Authentication is located in both MC and PHY.
While the MC node is down, the PHY node can continue to answer to authentication requests. However, updating membership and changing passwords will not be possible.
We may consider moving auth1 to DC for the outage.
CSC's DNS service is located in both MC and PHY.
We may consider moving the MC DNS node to DC, but this is not necessary to maintain services during the outage.
NOTE: The MC node is the master node, so we will need to ensure that the SOA record contains a long enough expiry time so the PHY doesn't stop serving zones.