Out of memory testing

This page describes how to use the test suite todo out of memory testing.

Building with OOM testing

Since OOM testing requires hooking into the malloc APIs, it is not enabled by default. The flag --enable-test-oom must be given to configure. When this is done the libvirt allocation APIs will have some hooks enabled.

$ ./configure --enable-test-oom

Basic OOM testing support

The first step in validating OOM usage is to run a test suite with full OOM testing enabled. This is done by setting the VIR_TEST_OOM=1 environment variable. The way this works is that it runs the test once normally to "prime" any static memory allocations. Then it runs it once more counting the total number of memory allocations. Then it runs it in a loop failing a different memory allocation each time. For every memory allocation failure triggered, it expects the test case to return an error. OOM testing is quite slow requiring each test case to be executed O(n) times, where 'n' is the total number of memory allocations. This results in a total number of memory allocations of '(n * (n + 1) ) / 2'

$ VIR_TEST_OOM=1 ./qemuxml2argvtest
 1) QEMU XML-2-ARGV minimal                                           ... OK
    Test OOM for nalloc=42 .......................................... OK
 2) QEMU XML-2-ARGV minimal-s390                                      ... OK
    Test OOM for nalloc=28 ............................ OK
 3) QEMU XML-2-ARGV machine-aliases1                                  ... OK
    Test OOM for nalloc=38 ...................................... OK
 4) QEMU XML-2-ARGV machine-aliases2                                  ... OK
    Test OOM for nalloc=38 ...................................... OK
 5) QEMU XML-2-ARGV machine-core-on                                   ... OK
    Test OOM for nalloc=37 ..................................... OK
...snip...

In this output, the first line shows the normal execution and the test number, and the second line shows the total number of memory allocations from that test case.

Tracking failures with valgrind

The test suite should obviously *not* crash during OOM testing. If it does crash, then to assist in tracking down the problem it is worth using valgrind and only running a single test case. For example, supposing test case 5 crashed. Then re-run the test with

$ VIR_TEST_OOM=1 VIR_TEST_RANGE=5 ../run valgrind ./qemuxml2argvtest
...snip...
 5) QEMU XML-2-ARGV machine-core-on                                   ... OK
    Test OOM for nalloc=37 ..................................... OK
...snip...
    

Valgrind should report the cause of the crash - for example a double free or use of uninitialized memory or NULL pointer access.

Tracking failures with stack traces

With some really difficult bugs valgrind is not sufficient to identify the cause. In this case, it is useful to identify the precise allocation which was failed, to allow the code path to the error to be traced. The VIR_TEST_OOM env variable can be given a range of memory allocations to test. So if a test case has 150 allocations, it can be told to only test allocation numbers 7-10. The VIR_TEST_OOM_TRACE variable can be used to print out stack traces.

$ VIR_TEST_OOM_TRACE=2 VIR_TEST_OOM=1:7-10 VIR_TEST_RANGE=5 \
    ../run valgrind ./qemuxml2argvtest
 5) QEMU XML-2-ARGV machine-core-on                                   ... OK
    Test OOM for nalloc=37 !virAllocN
/home/berrange/src/virt/libvirt/src/util/viralloc.c:180
virDomainDefParseXML
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:11786 (discriminator 1)
virDomainDefParseNode
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12677
virDomainDefParse
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12621
testCompareXMLToArgvFiles
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:107
virtTestRun
/home/berrange/src/virt/libvirt/tests/testutils.c:266
mymain
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:388 (discriminator 2)
virtTestMain
/home/berrange/src/virt/libvirt/tests/testutils.c:791
__libc_start_main
??:?
_start
??:?
!virAlloc
/home/berrange/src/virt/libvirt/src/util/viralloc.c:133
virDomainDiskDefParseXML
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:4790
virDomainDefParseXML
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:11797
virDomainDefParseNode
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12677
virDomainDefParse
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12621
testCompareXMLToArgvFiles
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:107
virtTestRun
/home/berrange/src/virt/libvirt/tests/testutils.c:266
mymain
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:388 (discriminator 2)
virtTestMain
/home/berrange/src/virt/libvirt/tests/testutils.c:791
__libc_start_main
??:?
_start
??:?
!virAllocN
/home/berrange/src/virt/libvirt/src/util/viralloc.c:180
virXPathNodeSet
/home/berrange/src/virt/libvirt/src/util/virxml.c:609
virDomainDefParseXML
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:11805
virDomainDefParseNode
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12677
virDomainDefParse
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12621
testCompareXMLToArgvFiles
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:107
virtTestRun
/home/berrange/src/virt/libvirt/tests/testutils.c:266
mymain
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:388 (discriminator 2)
virtTestMain
/home/berrange/src/virt/libvirt/tests/testutils.c:791
__libc_start_main
??:?
_start
??:?
!virAllocN
/home/berrange/src/virt/libvirt/src/util/viralloc.c:180
virDomainDefParseXML
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:11808 (discriminator 1)
virDomainDefParseNode
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12677
virDomainDefParse
/home/berrange/src/virt/libvirt/src/conf/domain_conf.c:12621
testCompareXMLToArgvFiles
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:107
virtTestRun
/home/berrange/src/virt/libvirt/tests/testutils.c:266
mymain
/home/berrange/src/virt/libvirt/tests/qemuxml2argvtest.c:388 (discriminator 2)
virtTestMain
/home/berrange/src/virt/libvirt/tests/testutils.c:791
__libc_start_main
??:?
_start
??:?
    

Non-crash related problems

Not all memory allocation bugs result in code crashing. Sometimes the code will be silently ignoring the allocation failure, resulting in incorrect data being produced. For example the XML parser may mistakenly treat an allocation failure as indicating that an XML attribute was not set in the input document. It is hard to identify these problems from the test suite automatically. For this, the test suites should be run with VIR_TEST_DEBUG=1 set and then stderr analysed for any unexpected data. For example, the XML conversion may show an embedded "(null)" literal, or the test suite might complain about missing elements / attributes in the actual vs expected data. These are all signs of bugs in OOM handling. In the future the OOM tests will be enhanced to validate that an error VIR_ERR_NO_MEMORY is returned for each allocation failed, rather than some other error.