Update TODO to take into account commit 11de40c
This commit is contained in:
parent
4f21e9ab46
commit
a4ae0ae270
|
@ -1,13 +1,9 @@
|
|||
Try changing the motd and saving the file. It'll reload. The log
|
||||
messages suggest that the server is dropping extant connection - as
|
||||
expected - but they don't actually drop. They just internally reboot
|
||||
as if each extant connection was fresh. The problem seems to be
|
||||
something like latency in shutdown-propagation:
|
||||
|
||||
- the new line-reader instance seems to be spawning before the old one exits (!!)
|
||||
|
||||
- but then also we see that the reconfiguration of the session
|
||||
listener is "too quick" and doesn't wait for the old connections to
|
||||
drop, so even if the line-reader instance correctly glitched out
|
||||
the old connection, we'd still have a zombie reborn session to deal
|
||||
with.
|
||||
expected - but it immediately comes back momentarily before going away
|
||||
properly. The session is able to reboot due to the glitching in
|
||||
assertion of the listen port *more quickly* than the latency of
|
||||
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.
|
||||
|
|
Loading…
Reference in New Issue