Instead of calculating the size of the cache repeatedly, which can take
a long time now that it can use a callback, instead cache the size and
update that on insertion and deletion.
This requires changing the cache descriptors to have two caches, one for
pending deferreds and the other for the actual values. There's no reason
to evict from the pending deferreds as they won't take up any more
memory.
The debug 'full_twisted_stacktraces' flag caused synapse to rewrite
twisted deferreds to always fire the callback on the next reactor tick.
This was to force the deferred to always store the stacktraces on
exceptions, and thus be more likely to have a full stacktrace when it
reaches the final error handlers and gets printed to the logs.
Dynamically rewriting things is generally bad, and in particular this
change violates assumptions of various bits of Twisted. This wouldn't
necessarily be so bad, but it turns out this option has been turned on
on some production servers.
Turning the option can cause e.g. #1778.
For now, lets just entirely nuke this option.
This only makes a difference for versions of ldap3 before 1.0, but a)
its best to be explicit and b) there are distributions that package
ancient versions for ldap3 (e.g. debian).
Allows delegating the password auth to an external module. This also
moves the LDAP auth to using this system, allowing it to be removed from
the synapse tree entirely in the future.
The only place that was observed was to set the profile. I've made it
so that the profile is set within store.register in the same transaction
that creates the user.
This required some slight changes to the registration code for upgrading
guest users, since it previously relied on the distributor swallowing errors
if the profile already existed.
We change it so that each cache has an individual CacheMetric, instead
of having one global CacheMetric. This means that when a cache tries to
increment a counter it does not need to go through so many indirections.
