> >Only pg_locks has it. And you can already get your VXID from there: > > SELECT virtualtransaction FROM pg_locks > WHERE pid = pg_backend_pid() LIMIT 1; While it is true that pg_locks contains the virtual transaction information, I believe there are strong technical reasons to expose this directly via a function.
Agreed.
First of all, querying pg_locks is expensive. By contrast, pg_current_vxact_id() is a practically free O(1) read from MyProc.
Yes, this is a significant advantage. The function simply reads from MyProc without any locking or iteration.
The %v log placeholder is the specific identifier for individual transaction executions (including read-only ones where no permanent XID is assigned). PIDs (%p) are session-scoped and too coarse for helping debug specific transactions in connection-pooled environments. This function allows applications to easily obtain the ID needed to correlate with server logs.
This is a compelling use case. In connection-pooled environments, correlating application-side logs with server logs by VXID is much more precise than using PIDs.
We already provide fast accessors for other identifiers like pg_backend_pid() and pg_current_xact_id(). Additionally, PostgreSQL often provides utility functions that overlap with other commands or views to improve developer experience (e.g., pg_notify() vs NOTIFY, pg_sleep() vs pg_sleep_for() vs pg_sleep_until()). It feels consistent to offer a simple accessor rather than requiring a complex query against a system view.
I agree. This follows established PostgreSQL patterns.
Best regards, Pavlo Golub
Additionally, the implementation is minimal (~20 lines), so the binary size impact is negligible. And since it's a leaf function called only when explicitly invoked by users, it has no impact on the main code path performance.
