codedge-llc / kadabra

89%

master: 87%

LAST BUILD ON BRANCH v0.3.5

branch: v0.3.5

CHANGE BRANCH

x

Reset

• v0.3.5
• better-queue
• chore/mix-cleanup
• ci/elixir-10-travis
• feat/keep-bin
• feat/less-process
• feat/params
• feat/socket
• fix/already-started
• fix/better-window-update
• fix/certifi
• fix/connection-sup
• fix/dual-settings
• fix/no-proc
• fix/proper-close
• fix/remove-warnings
• fix/temporary-restart
• flow
• master
• max-frame-size-fix
• new-sup
• refac/hpack
• socket-gen
• v0.3.6
• v0.3.7
• v0.3.8
• v0.4.0
• v0.4.1
• v0.4.2
• v0.4.3
• v0.4.4
• v0.4.5
• v0.5.0
• v0.6.0

Build # 190

Build Type

push

travis-ci

Committed by web-flow

Commit Message

fix: memory leak on connection crashes (#23)

* feat: better supervision structure

* chore: bumped kadabra to v0.3.5

Also: better test coverage

* feat: entirely new supervision strategy

Kadabra.Supervisor starts a ConnectionSupervisor and StreamSupervisor.
ConnSupervisor manages encoder/decoder tables and the Connection worker
with a :one_for_all restart strategy. StreamSupervisor manages the
streams with :one_for_one restart.

To keep the API the same, Kadabra.Supervisor.start_link returns the pid
of the supervisor itself. The Connection worker is keyed in the registry
by this supervisor pid. The user still makes calls with the returned
pid, but it's forwarded with a via_tuple to the proper worker. Unsure
if this might impact performance.

Possible TODO: Consider keying everything by supervisor pid instead of
the generated ref. While conflicts should be virtually nonexistent, this
would guarantee it.

* chore: update CHANGELOG

* fix: removed unused ConnectionSupervisor.start_connection

Run Details

333 of 373 relevant lines covered (89.28%)

10534.25 hits per line