Prior to this commit, a running VM could fail a ping check and be
destroyed. This causes issues when network hiccups occur or the machine
is performing a reboot.
A VM that is in a ready state will now be destroyed when handed back or
it hits the lifetime TTL.
An SSH check was added before moving a VM from pending to ready.
However, the result of that check did not matter and move_pending would
still be called. This moves the move_pending call to within the begin
block that holds the SSH check. If the check fails, then only
fail_pending will be called.
SSH should be available before a VM is moved from the 'pending' queue to
'ready'.
`check_ssh` should probably be a function in the tradition of DRY; I'm
going to hopefully follow up this PR with a `Vmpooler::Utility` library.
As we approach an "official" v1.0.0 of the API I'd like to remove some old
nested "ok" responses. These were left in as the Beaker vmpooler
hypervisor used them, but I long-ago patched that code and I think it's
time to deprecate these.
* rename the Redis token 'timestamp' var to 'created'
* update the Redis token 'last' var when token is successfully validataed
* expose the Redis token 'last' var in GET /token route
The return values from most redis calls inform the caller of whether a
key or hash value exists. Several exists() calls can be removed in
favor of this approach.
Updated spec tests to account for a removal of exists() and ismember()
calls in API tests.
Make a note in the logfile when a pool is detected to be empty.
Also:
- vmpooler__empty__<pool> Redis key to determine when to log
- lifetime/TTL checks moved to `_check_running_vm` method
- no longer pay attention to VMware-based 'host.runtime.bootTime'
This PR implements a bunch of other stuff to account for rspec testing:
- Thread creation and looping in `check_pool`
- Everything else in `_check_pool`
