Connection-level TempFilePages can hit 32,768G in 12.0.1.3298

[Breck_Carter]

Two very short-lived connections to a 12.0.1.3298 database with an 8K page size were recorded with TempFilePages = 4294967281 for one connection and 4294967287 for the other.

This is not easily reproduced (ok, it's not reproducible at all), but it sure is noticeable when peak values are tracked (e.g., by Foxhound).

Here are the recorded measurements...

connection_number,sample_recorded_at,TempFilePages
16,'2012-07-23 16:33:14.503',4294967287
16,'2012-07-23 16:33:24.718',4294967287
59,'2012-07-23 16:33:24.718',4294967281
16,'2012-07-23 16:33:35.804',4294967287
16,'2012-07-23 16:33:45.233',4294967287
16,'2012-07-23 16:33:55.549',4294967287
16,'2012-07-23 16:34:04.624',4294967287

[VolkerBarth]

And so your question is whether these values are correct and/or possible at all?

AFAIK, with 8K pages, the documented maximum dbspace size is 8.192 * 2^28 = 2^13 * 2^28 = 2^41, which is surely below the around 2^45 you have noticed...

[VolkerBarth]

...whereas my obvious question is: Where did you get these 32TB HDDs, Breck? 🙂

[Breck_Carter]

HDDs? Who said anything about HDDs? This is all in RAM!

Seriously, they are clearly bogus values, for two reasons: the sheer size of the values, and the fact they reached those sizes within 10 seconds of the each connection being established. The first might be conceivable at some point in the near future, but not the second.

[MarkCulp]

Converted to hex, the two values are 0xfffffff7 and 0xfffffff1. Very suspicious. Did you have any parallel queries executing at the time?