From 6703c5ef9a95b35f0d8ba6f616af093df2c05073 Mon Sep 17 00:00:00 2001 From: Tony Garnock-Jones Date: Sun, 6 Aug 2017 16:36:31 -0400 Subject: [PATCH] More notes on the ircd zombie session problem --- examples/ircd/TODO.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/examples/ircd/TODO.md b/examples/ircd/TODO.md index 961772a..b7ff9d6 100644 --- a/examples/ircd/TODO.md +++ b/examples/ircd/TODO.md @@ -7,3 +7,11 @@ teardown of the previous connection; so the new session-listener responds to the assertions from the old connection before the old connection has a chance to die. Of course, it *does* die (since commit 11de40c), but having that zombie reborn new session is annoying. + + - This is thorny. You'd think that having a session wait for its + line-reader to go would be enough, but the multiple nested + during/spawns creating the sessions mean that no matter how long + the old session instance sticks around, a new session will appear + before we're ready! ... maybe there's no way *at all* to + disambiguate old/new instances without, say, a unique + listener-socket identifier??