[syphar]

in certain load situations I saw:

where we seem to spend time in something that is not tracked in spans. My initial guess it the db connection pool.

So now we get:

• a span around Pool.get_async
• a small optimization in the source browser that frees the db connection earlier. might be a fix.
• a metric to track connection pool acquire duration

[syphar]

r? @GuillaumeGomez

[syphar]

short update:
the problem really was the connection pool:

since I increased max_connections in postgres two weeks ago, I just increased pool size and that helped.

The changes in this PR are still useful.

[syphar]

pool is full again:

[syphar]

( merged so I can solve / evaluate the production issuse)