It does some of what it says on the tin.
You can certainly test Oracle upgrades with it, replaying captures from 10g
onwards in your new test database.
Limitations include:
Licensing. You need either to use a snapshot standby system you already
have licences, or licence another test box. In theory if your client CPU is
likely to heavily load the database server, you also need to license
sufficient "client" boxes to run the workload replay clients.
If you have 16-core servers, an additional 16*US$50k? plus the RAT
supplement might mean the test needs heavy justification.
Captures are in a binary format. You cannot edit them. One limitation of
this is that you cannot replay to test into a new application, or
significant DDL changes to your existing system - though I haven't yet
tried it with versioning, which should allow regression testing on systems
with DDL changes.
With our workload, which was heavy and varied, there were a number of
transactions that simply would not run from RAT. The most severe cases of
these simply hung a synchronous replay, the only way to nearly complete it
was to run in async mode, which then introduces false errors due to
transactions running in an inconsistent order.
With a multiprocessing system, even if you had no hangs, there are
circumstances where a synchronous replay will experience different waits
than the original transaction on production, which might also cause
failure.
None of these are reason enough not to use it, if you have a use case -
which is basically regression testing a large part of your workload. Do not
underestimate the effort required to set a run up. On a large system for a
24 hour workload, bank on 5 working days to capture, restore the test
database, preprocess workload, set up and run the replay. If nothing goes
wrong, that is - if it's your first use, plan on a month.
There are free alternatives like James Morle's SIMora. But while it allows
you to create new transactions to test new deployments, unlike RAT, it too
has limitations. One being I think that it's not going to be easy to
capture 10046 traces for every transaction in your system if you want to
regression test an upgrade - the thing RAT's really designed for.
Regards,
JT
On Tue, 5 May 2015 at 23:46 Jeff Chirco <backseatdba@xxxxxxxxx> wrote:
Any of you use Oracle's Real Application Testing option? Do you love/hate
it? I am thinking of getting it but I haven't heard a lot of talk about
it. Please let me know your thoughts.
Thanks,
Jeff
