Re: [PATCH] core-image-sato-sdk: test image with 512M memory

[Kang Kai <Kai.Kang@windriver.com>]

On 2019/7/27 下午4:42, Kang Kai wrote:
> On 2019/7/27 上午5:40, richard.purdie@linuxfoundation.org wrote:
>> On Fri, 2019-07-26 at 05:23 -0400, kai.kang@windriver.com wrote:
>>> From: Kai Kang <kai.kang@windriver.com>
>>>
>>> When run do_testimage for core-image-sato-sdk, it fails to pass test
>>> case:
>>>
>>>> RESULTS - systemd.SystemdBasicTests.test_systemd_failed: FAILED
>>>> (0.43s)
>>> It is OOM issue and daemon rpc.statd is killed:
>>>
>>>>   [  531.306146] Out of memory: Kill process 193 (rpc.statd) score
>>>> 200 or sacrifice child
>>> Increase the memory of qemu to 512M to avoid such OOM issue.
>>>
>>> Signed-off-by: Kai Kang <kai.kang@windriver.com>
>>> ---
>>>   meta/recipes-sato/images/core-image-sato-sdk.bb | 1 +
>>>   1 file changed, 1 insertion(+)
>>>
>>> diff --git a/meta/recipes-sato/images/core-image-sato-sdk.bb
>>> b/meta/recipes-sato/images/core-image-sato-sdk.bb
>>> index d7cc52b52b..f7963d018e 100644
>>> --- a/meta/recipes-sato/images/core-image-sato-sdk.bb
>>> +++ b/meta/recipes-sato/images/core-image-sato-sdk.bb
>>> @@ -9,3 +9,4 @@ IMAGE_FEATURES += "dev-pkgs tools-sdk \
>>>     IMAGE_INSTALL += "kernel-devsrc"
>>>   +TEST_QEMUPARAMS = "-m 512"
>> Any idea what is using so much memory in the image when this happens?
>>
>> Its rather sad that we can't have NFS+systemd with 256MB memory...
>
> It caused by stap test case. I minimized the test cases to
>
> TEST_SUITES = "ping date ssh systemd stap kernelmodule gcc "
>
> which could reproduce the error.
>
> And it PASSes testimage that remove stap test from default TEST_SUITES:
>
> TEST_SUITES_remove = 'stap'
>
> But I can't reproduce the OOM failure to run stap tests manually.

Hi Richard,

The root cause of test case stap fails is available memory is too low. 
Compare systemd with sysvinit, memory status are listed:

systemd:
root@qemux86:~# free -h
               total        used        free      shared buff/cache   
available
Mem:          239Mi       120Mi        23Mi       8.0Mi 94Mi       100Mi
Swap:            0B          0B          0B

sysvinit:
root@qemux86:~# free -h
               total        used        free      shared buff/cache   
available
Mem:          239Mi        45Mi       111Mi       0.0Ki 82Mi       184Mi
Swap:            0B          0B          0B


With systemd, processes of rpc.statd and rpc.mountd take about less than 
80M memories.
And I compared with Debian 10, they take similar size of memories.

PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM TIME+ COMMAND
464 rpcuser   20   0   56316  51336   2192 S   0.0  20.9   0:00.09 rpc.statd
   186 root      20   0   30136  27024   2280 S   0.0  11.0 0:00.02 
rpc.mountd


Compare to sysvinit, they take only about 2M and 448K:

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM TIME+ COMMAND
   567 rpcuser   20   0    3444   2372   1792 S   0.0   1.0 0:00.00 
rpc.statd
   677 root      20   0    3700    448      0 S   0.0   0.2 0:00.00 
rpc.mountd


I didn't figure out the differences between these 2 ways to start the 
commands:

ExecStart=/usr/sbin/rpc.statd -F $STATD_OPTS

Vs.

test -x "$NFS_STATD" || NFS_STATD=/usr/sbin/rpc.statd
start-stop-daemon --start --exec "$NFS_STATD" --pidfile "$STATD_PID"


Regards,
Kai




>
>
> Regards,
> Kai
>
>>
>> Cheers,
>>
>> Richard
>>
>>
>

-- 
Kai Kang