We currently arbitrarily set a "one-size fits all" timeout in start-domain.
I suggest we cool-ify it like so:
1) Start with the usual timeout – 600 seconds or whatever.
2) For each of the very first 10 or 20 starts just note the actual start time in
a special file in the config directory or log directory.
- 10-20 allows them some time to make their system more complicated.
- Or we wait to see a pattern where the start times start to be +- 20% of each
3) Double that amount of time and make that the timeout
4) If it times out we say something smart like:
"Based on your domain's history it takes an average of 17 seconds to start. The
domain failed to start in 34 seconds. Check the server log for problems.
- we now reset to (1) and start collecting history again...
As a bonus this would be very useful for QE/Performance. We produce a file
automatically with startup times.