debian-mirror-gitlab/doc/administration/geo/disaster_recovery/planned_failover.md

194 lines
8.8 KiB
Markdown
Raw Normal View History

2020-06-23 00:09:42 +05:30
---
stage: Enablement
group: Geo
2021-02-22 17:27:13 +05:30
info: To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments
2020-06-23 00:09:42 +05:30
type: howto
---
2021-03-11 19:13:27 +05:30
# Disaster recovery for planned failover **(PREMIUM SELF)**
2019-07-31 22:56:46 +05:30
The primary use-case of Disaster Recovery is to ensure business continuity in
the event of unplanned outage, but it can be used in conjunction with a planned
failover to migrate your GitLab instance between regions without extended
downtime.
As replication between Geo nodes is asynchronous, a planned failover requires
a maintenance window in which updates to the **primary** node are blocked. The
length of this window is determined by your replication capacity - once the
**secondary** node is completely synchronized with the **primary** node, the failover can occur without
data loss.
This document assumes you already have a fully configured, working Geo setup.
2020-04-22 19:07:51 +05:30
Please read it and the [Disaster Recovery](index.md) failover
2019-07-31 22:56:46 +05:30
documentation in full before proceeding. Planned failover is a major operation,
and if performed incorrectly, there is a high risk of data loss. Consider
rehearsing the procedure until you are comfortable with the necessary steps and
have a high degree of confidence in being able to perform them accurately.
## Not all data is automatically replicated
2021-01-03 14:25:43 +05:30
If you are using any GitLab features that Geo [doesn't support](../index.md#limitations),
2019-07-31 22:56:46 +05:30
you must make separate provisions to ensure that the **secondary** node has an
up-to-date copy of any data associated with that feature. This may extend the
required scheduled maintenance period significantly.
A common strategy for keeping this period as short as possible for data stored
in files is to use `rsync` to transfer the data. An initial `rsync` can be
performed ahead of the maintenance window; subsequent `rsync`s (including a
final transfer inside the maintenance window) will then transfer only the
*changes* between the **primary** node and the **secondary** nodes.
Repository-centric strategies for using `rsync` effectively can be found in the
2020-04-22 19:07:51 +05:30
[moving repositories](../../operations/moving_repositories.md) documentation; these strategies can
2019-07-31 22:56:46 +05:30
be adapted for use with any other file-based data, such as GitLab Pages (to
be found in `/var/opt/gitlab/gitlab-rails/shared/pages` if using Omnibus).
2020-04-08 14:13:33 +05:30
## Preflight checks
2019-07-31 22:56:46 +05:30
2021-03-11 19:13:27 +05:30
NOTE:
In GitLab 13.7 and earlier, if you have a data type with zero items to sync,
this command reports `ERROR - Replication is not up-to-date` even if
replication is actually up-to-date. This bug was fixed in GitLab 13.8 and
later.
2020-07-28 23:09:34 +05:30
Run this command to list out all preflight checks and automatically check if replication and verification are complete before scheduling a planned failover to ensure the process will go smoothly:
```shell
gitlab-ctl promotion-preflight-checks
```
Each step is described in more detail below.
2019-07-31 22:56:46 +05:30
### Object storage
If you have a large GitLab installation or cannot tolerate downtime, consider
2020-04-22 19:07:51 +05:30
[migrating to Object Storage](../replication/object_storage.md) **before** scheduling a planned failover.
2019-07-31 22:56:46 +05:30
Doing so reduces both the length of the maintenance window, and the risk of data
loss as a result of a poorly executed planned failover.
2019-12-21 20:55:43 +05:30
In GitLab 12.4, you can optionally allow GitLab to manage replication of Object Storage for
2020-04-22 19:07:51 +05:30
**secondary** nodes. For more information, see [Object Storage replication](../replication/object_storage.md).
2019-12-21 20:55:43 +05:30
2019-07-31 22:56:46 +05:30
### Review the configuration of each **secondary** node
Database settings are automatically replicated to the **secondary** node, but the
`/etc/gitlab/gitlab.rb` file must be set up manually, and differs between
nodes. If features such as Mattermost, OAuth or LDAP integration are enabled
on the **primary** node but not the **secondary** node, they will be lost during failover.
Review the `/etc/gitlab/gitlab.rb` file for both nodes and ensure the **secondary** node
supports everything the **primary** node does **before** scheduling a planned failover.
### Run system checks
Run the following on both **primary** and **secondary** nodes:
2020-03-13 15:44:24 +05:30
```shell
2019-07-31 22:56:46 +05:30
gitlab-rake gitlab:check
gitlab-rake gitlab:geo:check
```
If any failures are reported on either node, they should be resolved **before**
scheduling a planned failover.
### Check that secrets match between nodes
The SSH host keys and `/etc/gitlab/gitlab-secrets.json` files should be
identical on all nodes. Check this by running the following on all nodes and
comparing the output:
2020-03-13 15:44:24 +05:30
```shell
2019-07-31 22:56:46 +05:30
sudo sha256sum /etc/ssh/ssh_host* /etc/gitlab/gitlab-secrets.json
```
If any files differ, replace the content on the **secondary** node with the
content from the **primary** node.
### Ensure Geo replication is up-to-date
The maintenance window won't end until Geo replication and verification is
completely finished. To keep the window as short as possible, you should
ensure these processes are close to 100% as possible during active use.
2021-04-17 20:07:23 +05:30
Go to the **Admin Area > Geo** dashboard on the **secondary** node to
2019-07-31 22:56:46 +05:30
review status. Replicated objects (shown in green) should be close to 100%,
and there should be no failures (shown in red). If a large proportion of
2020-05-24 23:13:21 +05:30
objects aren't yet replicated (shown in gray), consider giving the node more
2019-07-31 22:56:46 +05:30
time to complete
![Replication status](img/replication-status.png)
If any objects are failing to replicate, this should be investigated before
scheduling the maintenance window. Following a planned failover, anything that
failed to replicate will be **lost**.
2019-09-04 21:01:54 +05:30
You can use the [Geo status API](../../../api/geo_nodes.md#retrieve-project-sync-or-verification-failures-that-occurred-on-the-current-node) to review failed objects and
2019-07-31 22:56:46 +05:30
the reasons for failure.
A common cause of replication failures is the data being missing on the
**primary** node - you can resolve these failures by restoring the data from backup,
or removing references to the missing data.
### Verify the integrity of replicated data
2020-04-22 19:07:51 +05:30
This [content was moved to another location](background_verification.md).
2019-07-31 22:56:46 +05:30
### Notify users of scheduled maintenance
2020-10-24 23:57:45 +05:30
On the **primary** node, navigate to **Admin Area > Messages**, add a broadcast
message. You can check under **Admin Area > Geo** to estimate how long it
2019-07-31 22:56:46 +05:30
will take to finish syncing. An example message would be:
> A scheduled maintenance will take place at XX:XX UTC. We expect it to take
> less than 1 hour.
## Prevent updates to the **primary** node
2021-03-11 19:13:27 +05:30
To ensure that all data is replicated to a secondary site, updates (write requests) need to
be disabled on the primary site:
2019-07-31 22:56:46 +05:30
2021-03-11 19:13:27 +05:30
1. Enable [maintenance mode](../../maintenance_mode/index.md).
2019-07-31 22:56:46 +05:30
2020-04-08 14:13:33 +05:30
1. Disable non-Geo periodic background jobs on the **primary** node by navigating
2020-10-24 23:57:45 +05:30
to **Admin Area > Monitoring > Background Jobs > Cron**, pressing `Disable All`,
2019-07-31 22:56:46 +05:30
and then pressing `Enable` for the `geo_sidekiq_cron_config_worker` cron job.
This job will re-enable several other cron jobs that are essential for planned
failover to complete successfully.
## Finish replicating and verifying all data
1. If you are manually replicating any data not managed by Geo, trigger the
final replication process now.
2020-10-24 23:57:45 +05:30
1. On the **primary** node, navigate to **Admin Area > Monitoring > Background Jobs > Queues**
2019-07-31 22:56:46 +05:30
and wait for all queues except those with `geo` in the name to drop to 0.
These queues contain work that has been submitted by your users; failing over
before it is completed will cause the work to be lost.
2020-10-24 23:57:45 +05:30
1. On the **primary** node, navigate to **Admin Area > Geo** and wait for the
2019-07-31 22:56:46 +05:30
following conditions to be true of the **secondary** node you are failing over to:
2019-09-30 21:07:59 +05:30
- All replication meters to each 100% replicated, 0% failures.
- All verification meters reach 100% verified, 0% failures.
- Database replication lag is 0ms.
- The Geo log cursor is up to date (0 events behind).
2019-07-31 22:56:46 +05:30
2020-10-24 23:57:45 +05:30
1. On the **secondary** node, navigate to **Admin Area > Monitoring > Background Jobs > Queues**
2019-07-31 22:56:46 +05:30
and wait for all the `geo` queues to drop to 0 queued and 0 running jobs.
2020-04-22 19:07:51 +05:30
1. On the **secondary** node, use [these instructions](../../raketasks/check.md)
2020-04-08 14:13:33 +05:30
to verify the integrity of CI artifacts, LFS objects, and uploads in file
2019-07-31 22:56:46 +05:30
storage.
At this point, your **secondary** node will contain an up-to-date copy of everything the
**primary** node has, meaning nothing will be lost when you fail over.
## Promote the **secondary** node
2020-04-22 19:07:51 +05:30
Finally, follow the [Disaster Recovery docs](index.md) to promote the
2019-07-31 22:56:46 +05:30
**secondary** node to a **primary** node. This process will cause a brief outage on the **secondary** node, and users may need to log in again.
Once it is completed, the maintenance window is over! Your new **primary** node will now
begin to diverge from the old one. If problems do arise at this point, failing
2020-04-22 19:07:51 +05:30
back to the old **primary** node [is possible](bring_primary_back.md), but likely to result
2020-04-08 14:13:33 +05:30
in the loss of any data uploaded to the new **primary** in the meantime.
2019-07-31 22:56:46 +05:30
Don't forget to remove the broadcast message after failover is complete.