Here is a BDR problem we ran into recently:
A BDR group with a pair of nodes, N1 and N2, group is created on N1, N2 joins the group, so far so good
N2 departs/rejoins the group a couple of times, then ran into an issue, with the following symptom, after executing bdr.bdr_group_join() on N2 wrt N1:
a) N2 does not show up in N1’s bdr.bdr_nodes table (which now has only one record for N1 itself)
b) in N2’s bdr.bdr_nodes table, there is also only one record (for N2 itself), with node_status stuck at i
c) in N2’s log file, the following entries repeat (each time with a different PID):
< 2018-02-06 11:37:26.580 PST >LOG: worker process: bdr db: testdb (PID 526) exited with exit code 1
< 2018-02-06 11:37:31.622 PST >ERROR: previous init failed, manual cleanup is required
< 2018-02-06 11:37:31.622 PST >DETAIL: Found bdr.bdr_nodes entry for N2 (6519523095740643074,1,17222,) with state=i in remote bdr.bdr_nodes
< 2018-02-06 11:37:31.622 PST >HINT: Remove all replication identifiers and slots corresponding to this node from the init target node then drop and recreate this database and try again
The log message suggests that there is a record for N2 in N1’s bdr.bdr_nodes table, but as noted above, that table has only one record (for N1) stated.
Also note is that, each time before N2 joins, it’s thoroughly cleaned up (including bdr.remove_bdr_from_local_node(), drop and recreate database, etc.)
Now, no matter what I do on N1 to cleanup, the problem persists… the following have been tried, to no avail:
a) drop replication slots (if found)
b) 1. bdr.remove_bdr_from_local_node; 2. recreate BDR group
c) 1. drop bdr extension; 2. recreate bdr extension; 3. recreate BDR group
d) recycle postgres server after b.1) and c.1)
While dropping and creating database likely would fix the problem, doing so is unthinkable in a production environment.
Would appreciate thoughts and suggestions,
Thanks
