select count(*) from audit_log;
2652505123
From operational:
select nextval('audit_log_id_seq');
2752134460
If we exepect that the migration won't be run for two weeks, and we expect there to be 5 million rows added per day to the audit log between now and then, we get 14*5 = 70 million. 100 million to be safe. The sequence should be bumped forward 100 million from its current value of 2,752,134,460 giving us 2,852,134,460.
We are adding <300,000 IDs in our fixes, so this should be sufficient.
On average, we have a gap of about 20,000 IDs every weekday (about 7,000 on weekends). This gap means that we likely don't need to bump the ID forward at all...Should we, just to be safe?
From the logging server, we need to confirm that the sequence was reset on the same day (i.e., it was bumped back up to the number before the drop).
select id from audit_log_20130409 order by occurred_at desc limit 1;
553373632
select id from audit_log_20130521 order by occurred_at desc limit 1;
676407993
select id from audit_log_20130611 order by occurred_at desc limit 1;
735407034
select id from audit_log_20130625 order by occurred_at desc limit 1;
777476786
select id from audit_log_20140826 order by occurred_at desc limit 1;
1911234058
select id from audit_log_20140909 order by occurred_at desc limit 1;
1940192039
select id from audit_log_20140916 order by occurred_at desc limit 1;
1956379379