Add support for getting current timestamp in nano second precision.

[higuoxing]

This patch adds support for getting current timestamp in nano second precision by using clock_gettime() [1]. I didn't have a Windows machine so I didn't implement it for Windows. Feedbacks are welcome.

[1] https://man7.org/linux/man-pages/man2/clock_gettime.2.html

[fvannee]

This function as it is implemented is not IMMUTABLE, but it's VOLATILE. Normally, the Postgres now()/current_date etc. functions are neither of these, but are STABLE, which means that they return the same value across different calls in the same statement. I think any 'now' function that timestamp9 provides should be STABLE too. This means it likely requires a different implementation though.

[higuoxing]

Thanks for your comments! Sounds reasonable, I'm will revise it.

[higuoxing]

I'm using two static variables to 'cache' the timestamp inside a transaction. It's a little bit different from PostgreSQL's now(), since now() returns the first statement's timestamp inside a transaction to avoid extra cost of calling GetCurrentTimestamp()[1]. In this this approach, timestamp9_nsnow() returns the timestamp of the first timestamp9_nsnow() function call.

[1] https://github.com/postgres/postgres/blob/a8a44828a2e15c95a15549063205b779f7055f1c/src/backend/access/transam/xact.c#L2111

[fvannee]

Does this work in the case parallelism is enabled in queries? You can try using force_parallel_mode to test.

[higuoxing]

Nice catch! It seems that the current_xact_ts9_timestamp cannot be synced to parallel workers in my implementation like xactStartTimestamp. Maybe we can mark it with PARALLEL RESTRICTED?

https://www.postgresql.org/docs/current/parallel-safety.html