I updated my ‘never update’ coreos server and what could i tell you, it was a bad idea (update often or never). Disclaimer: If you have a running 0.4.x instance and want to update, this might not be the best guide for you. I made the mistake to update before backing up etcd, this guide is about recovering all your data from a “.ss” file, found in the old etcd data directory (found at /var/lib/etcd/).

Migrate your .ss file to a snapshot

First problem: You could not restore a cluster from a data directory in 0.4.x. If you try to start a 0.4.x server, this node can not join the cluster an refuses to work. Issue

Well then step up the version ladder, but slowly:

Get the etcd-v0.5.0-alpha.5-linux-amd64.tar.gz from the github releases.

Migrate your 0.4.x data directory to 0.5.x:

./etcd-migrate -data-dir="./data/etcd" -name="1426dada90d8448d816aac1358181a1a"

If you don’t know your effin etcd name anymore and the migrate tool does not help you, search the etcd/log file for something. If found mine because of a internal coreos entry.

Get access again

Now you have a proper .snap file inside etcd/snap/ something like 000...60.snap. Sadly i could not import this into etcd, but i thought at least i can run a 0.5.x server with it.

Adjust your ips:

./etcd --listen-client-urls= -data-dir ./data/etcd -name e1 --listen-peer-urls= --advertise-client-urls=

You might want to add -force-new-cluster.

At least you can access your data again:

./etcdctl --debug --peers ls /

Migrate to 2.0.0

Remove your etcd/wal directory. Download etcd 2.0.0 from the github release page.

./etcd-migrate -data-dir="./data/etcd" -name="1426dada90d8448d816aac1358181a1a"

Yes i had to do this again, maybe you could use the 2.0.0 migration without the 0.5.0, i don’t know.

Start your 2.0.0 server to test everything is fine:

./etcd --listen-client-urls= -data-dir ./data/etcd -name e1 --listen-peer-urls= --advertise-client-urls=

Transfer to 2.2

Transfer your data directory to your 2.2 server (to etcd2 directory and let the first start migrate itself. On Coreos you might simply want to start the service with the proper configuration:

systemctl start etcd2

Or manually but remember the set all cluster flags to avoid resetting the config again.

etcd2 --data-dir /var/lib/etcd2


Now, better create backups:

./etcdctl backup --data-dir ./data/etcd --backup-dir=./bkp