On Fri, 2007-03-23 at 17:16 -0600, Warren Little wrote: > My concern is that there were many more logfiles to be played > following "00000010000011A000000FD" > (ie 000000010000011E0000005C) yet it appears the recovery stop at that > point and called it good. > I would assume all WAL logs would be restored. I'm interested in your feedback here. How would you like it to have acted? The WAL file was clearly corrupt. 1. Don't continue and don't come up. Have the recovery fail. In order to bring the server up, we would have to restart recovery with an additional command to say "I note that my recovery has failed and would like recovery to come up at the last possible point." 2. Attempt to continue after we fail the CRC check. This is both dangerous and in many cases won't work either, since this is one of the normal ending points. 3. Continue after a CRC check, don't attempt to apply the records, just look at them to determine if they look correct. i.e. see if the CRC error applies to just that record 4. Add a command to ignore specific WAL records ignore_record = '11A/FD492B20' These may also not work very well at all, since many records depend upon previous data changes, so could quickly end in further errors. What would you suggest? -- Simon Riggs EnterpriseDB http://www.enterprisedb.com
