Just wanted to blog about some of the latest updates to dbqp. We just merged some interesting changes into Drizzle (just in time for the impending Fremont beta). In additional to general code cleanup / reorganization, we have the following goodies:
Randgen in the Drizzle tree
One of the biggest things is that the random query generator (aka randgen) is now part of the Drizzle tree. While I did some of the work here, the major drivers of this happening were Brian and Stewart:
- Brian makes a fair argument that the easier / more convenient it is to run a test, the greater the likelihood of it being run. Additional tools to install, etc = not so much. Having something right there and ready to go = win!
- Stewart is also a fan of convenience, lotsa testing, and working smarter, not harder. As a result, he did the initial legwork on merging the randgen. I do suspect there is still much for me to learn about properly bzr joining trees and whatnot, but we’ll get it right soon enough ; )
This doesn’t mean we won’t be contributing any changes we make back to the main randgen project / branch, it is strictly to facilitate more testing for Drizzle. As we already have our randgen tests packaged into dbqp-runnable suites, running these tests is even easier : )
Another request fulfilled in this update is the ability to use Stewart’s libeatmydata to speed up testing. By default, dbqp uses shared memory as a workdir, similar to mysql-test-run’s –mem option (this can be bypassed in dbqp with –no-shm, fyi). However, this isn’t always perfect or desirable to do.
An alternative is to use libeatmydata, which disables fsync() calls. As the name implies, you don’t want to use it if care about your data, but for general testing purposes, it can greatly speed up test execution.
If you have the library installed / on your machine, you can use it like so: ./dbqp –libeatmydata [–libeatmydata-path ] …
By default, libeatmydata-path is /usr/local/lib/libeatmydata.so (as if you used make install)
Multiple server types
IMHO, this is one of the coolest new tricks. dbqp can now handle more than just Drizzle servers / source! The ultimate idea is to allow tests that utilize more than one type / version of a server to have more interesting tests : ) This will be useful for scenarios like testing Drizzledump migration as we can feed in one (or more) MySQL servers and a Drizzle tree and make sure we can migrate data from all of them.
We also intend to utilize dbqp for testing a variety of Percona products, and it is kind of handy to be able to run the code you are testing ; ) I already have the tool running Percona / MySQL servers and have some randgen tests working:
$ ./dbqp.py --default_server_type=mysql --basedir=/percona-server/Percona-Server --mode=randgen
Setting --no-secure-file-priv=True for randgen usage...
20111013-163443 INFO Linking workdir /dbqp/workdir to /dev/shm/dbqp_workdir_pcrews_9dbc7e8a-2872-45a9-8a07-f347f6184246
20111013-163443 INFO Using mysql source tree:
20111013-163443 INFO basedir: /percona-server/Percona-Server
20111013-163443 INFO clientbindir: /percona-server/Percona-Server/client
20111013-163443 INFO testdir: /dbqp
20111013-163443 INFO server_version: 5.5.16-rel21.0
20111013-163443 INFO server_compile_os: Linux
20111013-163443 INFO server_platform: x86_64
20111013-163443 INFO server_comment: (Percona Server with XtraDB (GPL), Release rel21.0, Revision 188)
20111013-163443 INFO Using default-storage-engine: innodb
20111013-163443 INFO Using testing mode: randgen
20111013-163443 INFO Processing test suites...
20111013-163443 INFO Found 5 test(s) for execution
20111013-163443 INFO Creating 1 bot(s)
20111013-163449 INFO Taking clean db snapshot...
20111013-163452 INFO bot0 server:
20111013-163452 INFO NAME: s0
20111013-163452 INFO MASTER_PORT: 9307
20111013-163452 INFO SOCKET_FILE: /dbqp/workdir/bot0/s0/var/s0.sock
20111013-163452 INFO VARDIR: /dbqp/workdir/bot0/s0/var
20111013-163452 INFO STATUS: 1
20111013-163506 TEST NAME [ RESULT ] TIME (ms)
20111013-163506 main.blob [ pass ] 8624
20111013-163516 main.create_drop [ pass ] 2862
20111013-163524 main.many_indexes [ pass ] 1429
20111013-163547 main.optimizer_subquery [ pass ] 17153
20111013-163558 main.outer_join [ pass ] 4243
20111013-163558 INFO Test execution complete in 69 seconds
20111013-163558 INFO Summary report:
20111013-163558 INFO Executed 5/5 test cases, 100.00 percent
20111013-163558 INFO STATUS: PASS, 5/5 test cases, 100.00 percent executed
20111013-163558 INFO Spent 34 / 69 seconds on: TEST(s)
20111013-163558 INFO Test execution complete
20111013-163558 INFO Stopping all running servers...
Expect to see this up and running tests against Percona Server in the next week or so. I’ll be writing more about this soon.
Native / unittest mode
This hasn’t made it into the Drizzle tree yet. To ease merging the code with Percona Server / Xtrabackup, I’ve created a separate launchpad project. One of the things we needed was the ability to write complex tests directly. It is currently easy to plug new tools into dbqp, but we essentially needed a new tool for certain testing needs.
Our solution for this was to allow dbqp to run python unittest modules. We still have a bit of work to do before we have some demo tests ready, but we will be creating some expanded Xtrabackup tests using this system very soon. So far, it is turning out to be pretty neat:
./dbqp.py --default_server_type=mysql --basedir=/percona-server/Percona-Server --mode=native
20111013-190744 INFO Killing pid 1747 from /dbqp/workdir/bot0/s0/var/run/s0.pid
20111013-190744 INFO Linking workdir /dbqp/workdir to /dev/shm/dbqp_workdir_pcrews_9dbc7e8a-2872-45a9-8a07-f347f6184246
20111013-190744 INFO Using mysql source tree:
20111013-190744 INFO basedir: /percona-server/Percona-Server
20111013-190744 INFO clientbindir: /percona-server/Percona-Server/client
20111013-190744 INFO testdir: /dbqp
20111013-190744 INFO server_version: 5.5.16-rel21.0
20111013-190744 INFO server_compile_os: Linux
20111013-190744 INFO server_platform: x86_64
20111013-190744 INFO server_comment: (Percona Server with XtraDB (GPL), Release rel21.0, Revision 188)
20111013-190744 INFO Using default-storage-engine: innodb
20111013-190744 INFO Using testing mode: native
20111013-190744 INFO Processing test suites...
20111013-190744 INFO Found 1 test(s) for execution
20111013-190744 INFO Creating 1 bot(s)
20111013-190749 INFO Taking clean db snapshot...
20111013-190750 INFO bot0 server:
20111013-190750 INFO NAME: s0
20111013-190750 INFO MASTER_PORT: 9306
20111013-190750 INFO SOCKET_FILE: /dbqp/workdir/bot0/s0/var/s0.sock
20111013-190750 INFO VARDIR: /dbqp/workdir/bot0/s0/var
20111013-190750 INFO STATUS: 1
20111013-190756 TEST NAME [ RESULT ] TIME (ms)
20111013-190756 main.example_test [ pass ] 1
20111013-190756 test_choice (example_test.TestSequenceFunctions) ... ok
20111013-190756 test_sample (example_test.TestSequenceFunctions) ... ok
20111013-190756 test_shuffle (example_test.TestSequenceFunctions) ... ok
20111013-190756 Ran 3 tests in 0.000s
20111013-190756 INFO Test execution complete in 6 seconds
20111013-190756 INFO Summary report:
20111013-190756 INFO Executed 1/1 test cases, 100.00 percent
20111013-190756 INFO STATUS: PASS, 1/1 test cases, 100.00 percent executed
20111013-190756 INFO Spent 0 / 6 seconds on: TEST(s)
20111013-190756 INFO Test execution complete
20111013-190756 INFO Stopping all running servers...
This really only scratches the surface of what can happen, but I’ll be writing more in-depth articles on what kind of tricks we can pull off as the code gets more polished.
Three non-testing bits:
2) We are *this* close to Fremont beta being ready. The contributions and feedback have been most welcome. Any additional testing / etc are most appreciated.