- fix for gun worker termination in some circumstances
- pool for http clients (ex_aws, tzdata)
- default pool timeouts for gun
- gun retries on gun_down messages
- s3 upload timeout if streaming enabled
When gun shuts down due to the host being unreachable, the worker
process shuts down with the same shutdown reason since they are linked.
Gun doesn't have error tuples in it's shutdown reason though, so we need
to handle it in get_conn.
Closes#2008
This produced error log messages about GenServer termination
every time the connection was not open due to a timeout.
Instead we stop with `{:shutdown, <gun_error>}` since shutting down
when the connection can't be established is normal behavior.
- `verify_fun` is not useful now
- use `customize_check_hostname` (OTP 20+ so OK)
- `partial_chain` is useless as of OTP 21.1 (wasn't there, but hackney/..
uses it)
The numbers of the native time unit were so small the CRF was always 1,
making it an LRU. This commit switches the time to miliseconds and changes
the time delta multiplier to the one yielding mostly highest hit rates according
to the paper
`:retry_timeout` and `:retry` got removed because reconnecting on failure is
something the new pool intentionally doesn't do.
`:max_overflow` had to go in favor of `:max_waiting`, I didn't reuse the key because
the settings are very different in their behaviour.
`:checkin_timeout` got removed in favor of `:connection_acquisition_wait`,
I didn't reuse the key because the settings are somewhat different.
I didn't do any migrations/deprecation warnings/changelog entries because
these settings were never in stable.
While running this in production I noticed a number of ghost
processes with all their clients dead before they released the connection,
so let's track them to log it and remove them from clients
This patch refactors gun pooling to use Elixir process registry and
simplifies adapter option insertion.
Having the pool use process registry instead of a GenServer has a number of advantages:
- Simpler code: the initial implementation adds about half the lines of code it deletes
- Concurrency: unlike a GenServer, ETS-based registry can handle multiple checkout/checkin
requests at the same time
- Precise and easy idle connection clousure: current proposal for closing idle connections in
the GenServer-based pool needs to filter through all connections once a minute and compare their
last active time with closing time. With Elixir process registry this can be done
by just using `Process.send_after`/`Process.cancel_timer` in the worker process.
- Lower memory footprint: In my tests `gun-memory-leak` branch uses about 290mb on peak load (250 connections)
and 235mb on idle (5-10 connections). Registry-based pool uses 210mb on idle and 240mb on peak load
Haelwenn (lanodan) Monnier