runUnitTest: Close ZODB DB before shutdown
We are already closing Storage on shutdown (see nearby Storage.close call in the patch), but the DB handle was not closed. With classic ZODB it does not really matter in practice, because not closing DB is only leaking RAM resources and the program is anyway terminated soon. However with wendelin.core 2 things are different: in addition to ZODB storage server, there is also synthetic WCFS filesystem from which files are opened and memory-mmaped. In runUnitTest we start both ZODB and WCFS servers and we also shut down them both in the end. The filesystem server can be cleanly unmounted and shutdown only when there are no opened files left on it. Wendelin.core 2 client works by complementing each ZODB connection (zconn) with WCFS-level connection (wconn) to WCFS server. Those two zconn and wconn are kept in sync by wendelin.core 2 client logic: whenever zconn adjusts its view of the database, so does wconn. And whenever zconn is garbage collected, so does wconn is closed to release resources and close corresponding files opened on WCFS. In addition to garbage-collection, wconn is also closed when zconn.db - the ZODB DB handle via which zconn was created - is closed. This is needed to be able to reliably trigger freeing WCFS resources, because even after DB is closed, zconn can still stay alive forever being referenced from some python object - e.g. a frame or traceback or something else. The latter scenario actually happens during runUnitTest run. As the result it leads to inability to unmount and stop WCFS server cleanly: $ ./runUnitTest --load --save --with_wendelin_core -v erp5_core_test:testWendelinCore ... test (erp5.component.test.erp5_version.testWendelinCore.TestWendelinCoreBasic) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.105s OK F0618 19:05:46.359140 35468 wcfs/client/wcfs.cpp:486] CRITICAL: pinner: wcfs /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: wlink25: recvReq: link is down F0618 19:05:46.359173 35468 wcfs/client/wcfs.cpp:487] CRITICAL: wcfs server will likely kill us soon. CRITICAL: pinner: wcfs /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: wlink25: recvReq: link is down CRITICAL: wcfs server will likely kill us soon. Traceback (most recent call last): File ".../bin/runUnitTest", line 312, in <module> sys.exit(runUnitTest.main()) File ".../parts/erp5/Products/ERP5Type/tests/runUnitTest.py", line 926, in main run_only=run_only, File ".../parts/erp5/Products/ERP5Type/tests/runUnitTest.py", line 709, in runUnitTestList wcfs_server.stop() ... File ".../parts/wendelin.core/wcfs/__init__.py", line 543, in _fuse_unmount raise RuntimeError("%s\n(more details logged)" % emsg) RuntimeError: fuse_unmount /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: failed: fusermount: failed to unmount /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: Device or resource busy (more details logged) # logs 2021-06-18 19:05:45.978 INFO root wcfs: unmount/stop wcfs pid32981 @ /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2 2021-06-18 19:05:46.068 WARNING root fuse_unmount /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: failed: fusermount: failed to unmount /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2: Device or resource busy 2021-06-18 19:05:46.068 WARNING root # lsof /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2 2021-06-18 19:05:46.357 WARNING root COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME runUnitTe 32175 slapuser34 24r REG 0,48 111 4 /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2/.wcfs/zurl runUnitTe 32175 slapuser34 25u REG 0,48 0 7 /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2/head/watch runUnitTe 32175 slapuser34 26r REG 0,48 2097152 9 /dev/shm/wcfs/b53b61099c740b452b383db6df6dce4ad6d23ba2/head/bigfile/00000000000078b4 2021-06-18 19:05:46.358 WARNING root -> kill -TERM wcfs.go ... 2021-06-18 19:05:46.358 WARNING root -> abort FUSE connection ... I've debugged things a bit and even with kirr/ZODB@bbd03b3a ZODB connection stays alive being referenced from some frame objects. -> Fix this problem by explicitly closing ZODB DB nearby tests shutdown before call to wcfs_server.stop
Showing
Please register or sign in to comment