We had a server approach wraparound yesterday on a 9.3 box. I restarted
in single user mode and only one table was above
autovacuum_freeze_max_age. The dataset was ~750G. We left vacuum running
for 21hrs and it still did not complete. What was interesting is that
disk IO graph kept cycling, lots of reads for ~45min, then a few writes
for another 45min. I failed to VACUUM VERBOSE so there was little other
data to go on other than IO graph. We ended dropping the table and
lowering autovacuum_freeze_max_age for clean up. We've had similar sized
dataset complete vacuum in a few hours, but I've never seen this cycling
IO before: https://imagebin.ca/v/2eIdB2bjFBxq
Is this normal? It gave me very little faith that VACUUM would ever
complete.