Oracle 21 performs worse than previous versions and this
is causing some tool_httpsreplace jobs to, near consistently,
fail with timeouts.
This can be workaround in a number of ways:
1) Hacking tool_httpsreplace to detect when it's running a
behat test and avoid running it completely. We did something
like that for tool_customlang in the past. Ugly exception.
2) For Oracle jobs, set BEHAT_INCREASE_TIMEOUT to 2 or higher, that
will allow the whole behat run to have more time. Maybe too much
if only the tool_httpsreplace scenarios are affected. And requires
special configuration in sites running the tests.
3) Improve Oracle 21 performance. There are some remaining tests to
perform to achieve that, but it's extremely slow process, so it
won't happen soon.
4) Give more time to the failing tests using the step: "I mark this
test as slow setting a timeout factor of X". It applies to all
databases, but we are already using it, so no big problem. It's the
same than solution 2) but from within the tests instead of affecting
to the whole run.
From all the alternatives above, this patch implements 4), doubling
(from current 2 to 4), to give each test up to two minutes
(30 secs * 4 = 120 secs) with the new allowed timeout-factor.
Uses of the httpsreplace and customlang tools are known to be extremely
slow.
In our CI infrastructure on slwoer DBs, the customlang tool can take
over 90 seconds to load the page, whilst the httpsreplace tool can take
up to about 60 seconds.
These changes set appropriate timeout factors to increase the timeout
accordingly.
Eloy Lafuente (stronk7)