Upgrading Starter Deployments

The ArangoDB Starter supports an automated procedure to perform upgrades, including rolling upgrades of a Cluster setup.

The upgrade procedure of the Starter described in this Section can be used to upgrade to a new hotfix, or to perform an upgrade to a new minor version of ArangoDB. Please refer to the Upgrade Paths section for detailed information.

It is highly recommended to upgrade 3.6.x and 3.7.x deployments using at least the starter version 0.15.0-1 because of a technical problem, see Technical Alert #6.

Upgrade Scenarios

The following four cases are possible:

  1. You have installed via an installation package (e.g. a .deb or .rpm package) and you will upgrade this installation using again an installation package (e.g. a .deb or .rpm).
  2. You have installed via the .tar.gz distribution and you will upgrade this installation using again a .tar.gz distribution.
  3. You have installed via an installation package (e.g. a .deb or .rpm package) and you will upgrade this installation using a .tar.gz distribution.
  4. You have installed via the .tar.gz distribution and you will upgrade this installation using an installation package (e.g. a .deb or .rpm package).

Cases 1. and 2. are more common, though cases 3. and 4. are also possible.

Upgrade Procedure

The following procedure has to be executed on every ArangoDB Starter instance. It is assumed that a Starter deployment with mode single, activefailover or cluster is running.

Install the new ArangoDB version binary

Installing the new ArangoDB version binary also includes the latest ArangoDB Starter binary, which is necessary to perform the rolling upgrade.

The first step is to install the new ArangoDB package.

Note: you do not have to stop the Starter processes before upgrading it.

For example, if you want to upgrade to 3.7.13 on Debian or Ubuntu, either call

  1. apt install arangodb=3.7.13

(apt-get on older versions) if you have added the ArangoDB repository. Or install a specific package using

  1. dpkg -i arangodb3-3.7.13-1_amd64.deb

after you have downloaded the corresponding file from www.arangodb.com/download/.

If you are using the .tar.gz distribution, you can simply extract the new archive in a different location and keep the old installation where it is. Note that this does not launch a standalone instance, so the following section can be skipped in this case.

Stop the Standalone Instance

As the package will automatically start the standalone instance, you might want to stop it now, as otherwise this standalone instance that is started on your machine can create some confusion later. As you are using the Starter you do not need this standalone instance, and you can hence stop it:

  1. service arangodb3 stop

Also, you might want to remove the standalone instance from the default runlevels to prevent it to start on the next reboot of your machine. How this is done depends on your distribution and init system. For example, on older Debian and Ubuntu systems using a SystemV-compatible init, you can use:

  1. update-rc.d -f arangodb3 remove

Stop the Starter without stopping the ArangoDB Server processes

Now all the Starter (arangodb) processes have to be stopped. Please note that no arangod processes should be stopped!

In order to stop the arangodb processes, leaving the arangod processes they have started up and running (as we want for a rolling upgrade), we will need to use a command like kill -9.

When using SystemD as supervisor, make sure that the unit file contains KillMode=process (see systemd.kill documentation). Otherwise kill -9 will not just kill the respective arangodb starter process, but also the arangod server processes it started because of the default setting KillMode=control-group.

  1. kill -9 <pid-of-starter>

The PID associated to your Starter can be checked using a command like ps:

  1. ps -C arangodb -fww

The output of the command above does not only show the PIDs of all arangodb processes but also the used commands, which can be useful for the following restart of all arangodb processes.

The output below is from a test machine where three instances of a Starter are running locally. In a more production-like scenario, you will find only one instance of arangodb running:

  1. ps -C arangodb -fww
  2. UID PID PPID C STIME TTY TIME CMD
  3. max 29419 3684 0 11:46 pts/1 00:00:00 arangodb --starter.data-dir=./db1
  4. max 29504 3695 0 11:46 pts/2 00:00:00 arangodb --starter.data-dir=./db2 --starter.join 127.0.0.1
  5. max 29513 3898 0 11:46 pts/4 00:00:00 arangodb --starter.data-dir=./db3 --starter.join 127.0.0.1

You can use pstree to inspect the arangod server instances launched by one of these starters:

  1. pstree -Tp 29419
  2. arangodb(29419)─┬─arangod(30201)
  3. ├─arangod(30202)
  4. └─arangod(30217)

Restart the Starter

When using a supervisor like SystemD, this will happen automatically. In case the Starter was initiated manually, the arangodb processes have to be restarted manually with the same command that has been used before.

You can inspect which processes belong to a starter instance using the pstree command (see above).

If you are using the .tar.gz distribution (only available from v3.4.0), your new version of the executable might be located in a different directory. Make sure that you now start the new Starter executable (bin/arangodb) in the new installation place. If you are using a supervisor like SystemD, you might have to adjust the path to the executable in the service description to the new location. Do this before you kill -9 the Starter or else the old version will be restarted in this case. If you forgot, simply do the kill -9 again.

After you stopped the Starter make sure the arangod processes it spawned are still running; they should be re-parented to the systemd or init process:

  1. ps -f -p 30201,30202,30217
  2. UID PID PPID C STIME TTY TIME CMD
  3. root 30201 1 0 13:02 pts/45 00:01:09 usr/sbin/arangod ...
  4. root 30202 1 0 13:02 pts/45 00:00:55 usr/sbin/arangod ...
  5. root 30217 1 0 13:02 pts/45 00:01:11 usr/sbin/arangod ...

If not, rollback to the old version and restart that Starter, or the subsequent upgrade procedure will fail.

After you have successfully restarted the Starter you will find yourself in the following situation:

  • The Starter is up and running, and it is on the new version
  • The ArangoDB Server processes are up and running, and they are still on the old version

Start the upgrade process of all arangod & arangosync servers

Run the following command for any of the starter endpoints (e.g. http://localhost:8528) to upgrade the entire cluster:

  1. arangodb upgrade --starter.endpoint=<endpoint-of-a-starter>

If you have connected clusters across multiple datacenter (DC2DC deployment), then you need to update each of the clusters.

Deployment mode single

For deployment mode single, the arangodb upgrade command will:

  • Restart the single server with an additional --database.auto-upgrade=true argument. The server will perform the auto-upgrade and then stop. After that the Starter will automatically restart it with its normal arguments.

The arangodb upgrade command will complete right away. Inspect the log of the Starter to know when the upgrade has finished.

Deployment mode activefailover or cluster

The Starters will now perform an initial check that upgrading is possible and when that all succeeds, create an upgrade plan. This plan is then executed by every Starter.

The arangodb upgrade command will show the progress of the upgrade and stop when the upgrade has either finished successfully or finished with an error.

Uninstall old package

This step is required in the cases 2., 3. and 4. only. It is not required in case 1., see Upgrade Scenarios above.

After verifying your upgraded ArangoDB system is working, you can remove the old package. This can be done in different ways, depending on the case you are:

  • Cases 2. and 4.: just remove the old directory created by the .tar.gz (assumes your --starter.data-dir is located outside of this directory - which is a recommended approach).
  • Case 3.: just remove the old package by running the corresponding uninstallation command (the exact command depends on whether you are using a .deb or .rmp package and it is assumed that your --starter.data-dir is located outside of the standard directories created by the installation package - which is a recommended approach).

Retrying a failed upgrade

When an upgrade plan (in deployment mode activefailover or cluster) has failed, it can be retried.

To retry, run:

  1. arangodb retry upgrade --starter.endpoint=<endpoint-of-a-starter>

The --starter.endpoint option can be set to the endpoint of any of the starters, e.g. http://localhost:8528.

Aborting an upgrade

When an upgrade plan (in deployment mode activefailover or cluster) is in progress or has failed, it can be aborted.

To abort, run:

  1. arangodb abort upgrade --starter.endpoint=<endpoint-of-a-starter>

The --starter.endpoint option can be set to the endpoint of any of the starters, e.g. http://localhost:8528.

Note that an abort does not stop all upgrade processes immediately. If an arangod or arangosync server is being upgraded when the abort was issued, this upgrade will be finished. Remaining servers will not be upgraded.