corrupted introducer.furl causes subtle startup breakage #371

New Issue

warner · 2008-03-28T18:59:40Z

warner commented

2008-03-28 18:59:40 +00:00

Timothy Fitz discovered last night at the hackfest that a corrupt
BASEDIR/introducer.furl (caused by a broken link on our TestGrid page) causes
the node to throw an exception partway through the startup process, leaving
it in a funny state. Basically the web server got attached, but the client
instance didn't get attached (which happens in startService). This made the
welcome page display a traceback complaining about being unable to adapt the
nevow context to the IClient interface.

We had some code in place (at rob's prompting) to detect exceptions during
startup and then do sys.exit(). I suspect that there's an error path that
bypasses this. Note that the exception did show up in twistd.log, but the
node ran anyways.

The #71 "node probably started" feedback should cover problems like this: if
node startup fails in this way, the 'tahoe start' command should not
display success.

Timothy Fitz discovered last night at the hackfest that a corrupt BASEDIR/introducer.furl (caused by a broken link on our [TestGrid](wiki/TestGrid) page) causes the node to throw an exception partway through the startup process, leaving it in a funny state. Basically the web server got attached, but the client instance didn't get attached (which happens in startService). This made the welcome page display a traceback complaining about being unable to adapt the nevow context to the `IClient` interface. We had some code in place (at rob's prompting) to detect exceptions during startup and then do sys.exit(). I suspect that there's an error path that bypasses this. Note that the exception *did* show up in twistd.log, but the node ran anyways. The #71 "node probably started" feedback should cover problems like this: if node startup fails in this way, the 'tahoe start' command should *not* display success.