706678 - fix server-full `make test` to clean up/use temp names for sqlite files and log files

Status: VERIFIED FIXED

(Cloud Services Graveyard :: Server: Sync, defect, P4)

Description

[John Morrison [:jrgm]]

fix server-full `make test` to clean up/use temp names for sqlite files and log files 
-> jrgm

Comment 2

[Ryan Kelly [:rfkelly]]

The changes in Bug 740703 should resolve this problem, by ensuring that comprehensive database cleanup code is run for every test.  This is much easier than using temp filenames for the database files, since the filenames are pulled out of static config files.

John/James, can you please try the usual `make build; make test` with server-syncstorage master and confirm that it cleans up all the files you expect it to clean up?

Comment 3

[James Bonacci [:jbonacci]]

Marking Resolved for :rfkelly.

Comment 4

[James Bonacci [:jbonacci]]

Actually, I am seeing a lot of failures now on a fresh upgrade of server-full on qa1. But this may be related to other issues with that install.

See http://jbonacci.pastebin.mozilla.org/1572351

Comment 5

[Ryan Kelly [:rfkelly]]

This looks like an old, incompatible database file is still present in /tmp/tests.db.  If you remove it manually, it should work and the database file should no longer be left behind at the end.

(Note that this patch only ensures to clean up database created while running the tests, not any that might already exist on the system)

Comment 6

[Ryan Kelly [:rfkelly]]

Err, since this bug is explicitly about server-full, I should backport this into the sync1.1 codebase for completeness.

Comment 7

[James Bonacci [:jbonacci]]

Noting that this is all for sync 1.1:

OK, so on qa1, the following files in /tmp were either 1. created/updated by make build, 2. created/updated by make test, or some combination of both (since they all had new time/date stamps:
another-test-host.db  reset.db  some-test-host.db  sync.db  tests.db
sync-error.log

I cleared these out and ran "make test" again and got a clean run.
But once again, I see these files left behind in /tmp:
another-test-host.db  reset.db  some-test-host.db  sync.db  sync-error.log  tests.db
All with very current date/time stamps (for the current run of "make test")

On a clean install on rhel6.build.mtv1:
Right after "make build", I see these files in /tmp:
account.db  reset.db  sync-error.log  tests.db

Fixes (if working for 2.0, should be back-ported to 1.1)

Comment 8

[James Bonacci [:jbonacci]]

Moving on to sync 2.0:
After two calls to "make build" (because of that dang PasteDeploy bug), I see no temp files left in /tmp

After a clean run of "make test" I verified no stray files left in /tmp.
And not seeing anything newer than nosetests.xml in the server-syncstorage tree.

Comment 9

[Ryan Kelly [:rfkelly]]

Attachment: "backport" to sync1.1

Duplicating the additional cleanup steps into sync1.1 codebase.  Not really a backport as the sync2.0 version uses our new testing framework, but it should produce the same result.

Comment 10

[James Bonacci [:jbonacci]]

Let me know when this is ready (and set to Resolved/Fixed). I will check against an install of Sync 1.1 on qa1.

Comment 11

[Ryan Kelly [:rfkelly]]

Attachment: patch to remove tempfiles for server-core

Comment 12

[Ryan Kelly [:rfkelly]]

Attachment: patch to remove tempfiles for server-reg

I've also done similar changes for server-core and server-reg.  With this in place, `make build` and `make test` for server-full should no longer leave stray tempfiles.  This is good both for third-party installers and our lovely QA folks.

Comment 13

[Ryan Kelly [:rfkelly]]

http://hg.mozilla.org/services/server-reg/rev/5efc3c7c37ff
http://hg.mozilla.org/services/server-reg/rev/5efc3c7c37ff

The default `make build` for server-full builds things off the production channel, so it won't pick up these changes straight away.  The following, however, should run without leaving behind any tempfiles:

    make build CHANNEL=dev
    make test

Comment 14

[James Bonacci [:jbonacci]]

OK. I can verify this on qa1, sync 1.1 using the channel param.

Comment 15

[James Bonacci [:jbonacci]]

OK, so I backed up the current server-full.
I deleted leftover files in /tmp: the .db files and the one .log file
Did the usual install process, with this change:
$ make build CHANNEL=dev

see this in STDOUT:
...etc...
Cleaning up...
bin/buildapp -c dev -i http://pypi.python.org/simple server-core,server-reg,server-storage
The current channel is dev.
Building the app 
...etc...

Run make test
$ make test
bin/nosetests -s --with-xunit deps/server-core/services/tests deps/server-reg/syncreg/tests deps/server-storage/syncstorage/tests
...............................................................................S.S..S........................./opt/sync/server-full/deps/server-reg/syncreg/tests/functional/support.py:92: DeprecationWarning: Method SQLUser.delete_user should be passed a dict of credentials, not a raw password
  self.auth.delete_user(self.user, self.password)
./opt/sync/server-full/deps/server-reg/syncreg/controllers/user.py:259: DeprecationWarning: Method SQLUser.update_field should be passed a dict of credentials, not a raw password
  'mail', email):
./opt/sync/server-full/deps/server-reg/syncreg/controllers/user.py:309: DeprecationWarning: Method SQLUser.update_password should be passed a dict of credentials, not a raw password
  new_password):
../opt/sync/server-full/deps/server-reg/syncreg/tests/functional/test_user.py:349: DeprecationWarning: Method SQLUser.delete_user should be passed a dict of credentials, not a raw password
  self.auth.delete_user(User(name), 'x' * 9)
./opt/sync/server-full/deps/server-reg/syncreg/controllers/user.py:403: DeprecationWarning: Method SQLUser.delete_user should be passed a dict of credentials, not a raw password
  res = self.auth.delete_user(request.user, request.user_password)
........./opt/sync/server-full/deps/server-reg/syncreg/tests/functional/test_user.py:537: DeprecationWarning: Method SQLUser.delete_user should be passed a dict of credentials, not a raw password
  self.auth.delete_user(User(name), 'x' * 9)
.......................................................
----------------------------------------------------------------------
Ran 179 tests in 53.882s

OK (SKIP=3)

First, there are no *.db or *.log files in /tmp, so that is fixed.

Let me know if the above deprecation warnings and skipped tests are expected given my configuration which does use MySQL setup as shown in my server-full/etc/sync.conf file (rather than using the default sqlite3 config).
I have user "mysql.user" with a password accessing the sync.db database.

Comment 16

[Ryan Kelly [:rfkelly]]

The deprecation warnings are expected, and I get the same three skips on my machine due to missing libs or services.  The skipped tests are:

  * testing the LDAP user backend
  * testing the SReg user backend
  * testing some mocked backend errors

So I'm happy to declare this the expected output.

Comment 17

[James Bonacci [:jbonacci]]

Awesome. Will roll back to Prod channel at some point on qa1. But, for now, calling this one Verified.

Comment 18

[Ryan Kelly [:rfkelly]]

Backported to 1.10 release branch: http://hg.mozilla.org/services/server-storage/rev/2136078ea1d0

Comment 19

[James Bonacci [:jbonacci]]

Just re-re-confirmed that the backported fix to Sync 1.1 is not working correctly.
Even with a Dev branch build and some specific RPM packages chosen, the following two .db files are still getting created in /tmp:
/tmp/sync.db  /tmp/test.db

See the long-lasting pastebin of our IRC conversation for more information:
http://jbonacci.pastebin.mozilla.org/1608388


STRs:
$ hg clone https://hg.mozilla.org/services/server-full
$ cd server-full
$ make build CHANNEL=dev
$ make build SERVER_STORAGE=rpm-1.10-2 SERVER_REG=rpm-1.0-1 SERVER_CORE=rpm-2.6-5
$ make test
Gives two errors around this:
ImportError: No module named emailer
for syncreg
and the two files in /tmp

Comment 20

[Ryan Kelly [:rfkelly]]

I've made some fresh tags so that the fix gets pulled in by default.  On my box, the following build and run without leaving any database files behind:

Building on default (prod) channel:
$ hg clone https://hg.mozilla.org/services/server-full
$ cd server-full
$ make build
$ make test


Building on the development channel:
$ hg clone https://hg.mozilla.org/services/server-full
$ cd server-full
$ make build CHANNEL=dev
$ make test


To get this working with the deployed version of server-storage, we would also need to backport the changes into server-core 2.6 branch.  Is it worthwhile to do so?

Comment 21

[Toby Elliott [:telliott]]

I don't think it's worthwhile. Assumption is that external people will be on the trunk, and we don't use sqlite or care about log file location all that much.

Comment 22

[Ryan Kelly [:rfkelly]]

OK, marking this "resolved/as-fixed-as-we're-going-to-make-it".

Comment 23

[James Bonacci [:jbonacci]]

OK. I can mark this as Verified/Fixed Enough.