This is critical for LP: we're currently unable to deploy to the
codehosting service again, until this is fixed and pushed out to SRUs.
The logic being:
- we can't reasonably implement a server side 'please disconnect now'
trigger to the bzr server if it will cause all current clients to
error
- we can't deploy to the service if it needs an admin to baby sit it for hours
- and we can't set the idle time low enough to timeout on CI servers
if the timeout causes clients to error
- and so with no timeout and no explicit 'please stop now' deploys to
that service always end up having to interrupt a connection that is in
an unknown state.
Its also critical path for the faster ssh daemon stuff, because until
we can deploy reliably we can't half-deploy and experiment with
confidence.
This is critical for LP: we're currently unable to deploy to the
codehosting service again, until this is fixed and pushed out to SRUs.
The logic being:
- we can't reasonably implement a server side 'please disconnect now'
trigger to the bzr server if it will cause all current clients to
error
- we can't deploy to the service if it needs an admin to baby sit it for hours
- and we can't set the idle time low enough to timeout on CI servers
if the timeout causes clients to error
- and so with no timeout and no explicit 'please stop now' deploys to
that service always end up having to interrupt a connection that is in
an unknown state.
Its also critical path for the faster ssh daemon stuff, because until
we can deploy reliably we can't half-deploy and experiment with
confidence.
-Rob