Galera attempts to recover primary view if both pc.recovery and --wsrep-new-cluster are set

[GeoffMontee]

When bootstrapping a new cluster with --wsrep-new-cluster, it doesn't make sense to recover an old primary view from gvwstate.dat, since the user is explicitly asking to bootstrap a new cluster. However, in that situation, Galera does still attempt to recover the old primary view if pc.recovery is set.

We can show this by reproducing issue #354 and then bootstrapping. For example:

[ec2-user@ip-172-31-31-73 ~]$ sudo cp /dev/null /var/lib/mysql/gvwstate.dat
[ec2-user@ip-172-31-31-73 ~]$ sudo service mysql bootstrap
Bootstrapping the cluster.. Starting MySQL... ERROR!

And then the error log shows:

150929 12:02:36 [Note] WSREP: 'wsrep-new-cluster' option used, bootstrapping the cluster
150929 12:02:36 [Note] WSREP: Setting initial position to 3b43c7c9-45c3-11e5-b1d7-be55ee2e2933:10
150929 12:02:36 [Note] WSREP: protonet asio version 0
150929 12:02:36 [Note] WSREP: Using CRC-32C for message checksums.
150929 12:02:36 [Note] WSREP: backend: asio
150929 12:02:36 [Note] WSREP: restore pc from disk successfully
150929 12:02:36 [Note] WSREP: GMCast version 0
150929 12:02:36 [Note] WSREP: (00000000, 'tcp://0.0.0.0:4567') listening at tcp://0.0.0.0:4567
150929 12:02:36 [Note] WSREP: (00000000, 'tcp://0.0.0.0:4567') multicast: , ttl: 1
150929 12:02:36 [ERROR] WSREP: failed to open gcomm backend connection: 131: invalid UUID: 00000000 (FATAL)
         at gcomm/src/pc.cpp:PC():309
150929 12:02:36 [ERROR] WSREP: gcs/src/gcs_core.cpp:long int gcs_core_open(gcs_core_t*, const char*, const char*, bool)():206: Failed to open backend connection: -131 (State not recoverable)
150929 12:02:36 [ERROR] WSREP: gcs/src/gcs.cpp:long int gcs_open(gcs_conn_t*, const char*, const char*, bool)():1379: Failed to open channel 'my_wsrep_cluster' at 'gcomm://172.31.31.72,172.31.31.73,172.31.20.230': -131 (State not recoverable)
150929 12:02:36 [ERROR] WSREP: gcs connect failed: State not recoverable
150929 12:02:36 [ERROR] WSREP: wsrep::connect() failed: 7
150929 12:02:36 [ERROR] Aborting

I believe this is fixed in pull request #332. See here: https://github.com/nirbhayc/galera/commit/70559dc1d8d3f88dc08fbe38a387b3507bba3410#diff-f74f5be8fb18cad3dd5bd7049b19dc8eR98

[anthonyoteri]

I'm running into the same issue at one of our customer's sites. Is there any workaround / solution which can be performed which will allow our customer to recover from this?