Thanks. For the post above, this isn't a problem with temdb data, but the tempdb log.

Quick update. Yesterday, I monitored the log usage all day and found it did work as Microsoft designed, with the log filling up to 70% and then truncating. This was frustrating in a way as I meant I wasn't replicating what had been happening over the last few days.

However, I came into the office today and found the log had grown to 23GB overnight and was at 99% full. This time though I managed to find that an active transaction was actually running. Unfortunately, I couldn't identify what the query was being executed. Sp_Whoisactive just had the query text as "Begin Transaction". (I'll look now to see if there is a better way to get the query text from a session.) However, it does point me to the application which is causing this, and does prove that it isn't a bug or something weird happening in TempDB. I ended up killing the session and the log usage then truncated and went to 0.2% full as expected.

So, I can now investigate what in the application is causing the log to grow so big.

Thanks all for your help.