So I've been working on deleting a node with some 40M objects (80M, if you count copy pools). I kicked it off yesterday afternoon, and by this morning it had deleted a fine total of 128K objects. Not auspicious.
The server instance was active (though not really busy), but it's essentially constantly active, so I didn't really pay much attention. But a crash in a different instance (ran out of log space, was trying to DB backup, but tapes were all occupied, etc.) made me want to quiesce not just the instance, but the whole system image. In an attempt to do that, I slew the filespace delete, and all other work on the server. Yet the thing was still occupying a whole CPU, and doing very substantial IO. I want to emphasize, no sessions except for my admin session doing the queries, and no processes. Anyone ever seen something like that? I restarted the server instance and the behavior hasn't yet repeated. - Allen S. Rout
