It's a good idea to split a big patch into several smaller ones.
But you've already implemented these features - you could provide them
as sequential small patches (i.e. v13-0002-guc-context-propagation.patch and so on)
Great job! I'm both hands for committing your patch set.
On Fri, Jan 26, 2024 at 1:17 PM Anthonin Bonnefoy <[email protected]> wrote:
Hi,
The last CFbot failure was on pg_upgrade/002_pg_upgrade and doesn't seem related to the patch.
# Failed test 'regression tests pass' # at C:/cirrus/src/bin/pg_upgrade/t/002_pg_upgrade.pl line 212. # got: '256' # expected: '0' # Looks like you failed 1 test of 18.
Given that the patch grew a bit in complexity and features, I've tried to reduce it to a minimum to make reviewing easier. The goal is to keep the scope focused for a first version while additional features and changes could be handled afterwards.
For this version: - Trace context can be propagated through SQLCommenter - Sampling decision uses either a global sampling rate or on SQLCommenter's sampled flag - We generate spans for Planner, ExecutorRun and ExecutorFinish
What was put on hold for now: - Generate spans from planstate. This means we don't need the change in instrument.h to track the start of a node for the first version. - Generate spans for parallel processes - Using a GUC variable to propagate trace context - Filtering sampling on queryId
With this, the current patch provides the core to: - Generate, store and export spans with basic buffer management (drop old spans when full or drop new spans when full) - Keep variable strings in a separate file (a la pg_stat_statements) - Track tracing state and top spans across nested execution levels
