Hi Eddie,
ujson is one of the few packages that has no wheel, so it is being recompiled each time. It might be the case that some change to gcc and/or python-dev parts caused that. The issue looks like an edge case of c99 spec non-conformity so valid compiles may break on it (or work - just that there is no guarantee).
-yoctozepto
czw., 30 sty 2020 o 09:06 Eddie Yen missile0407@gmail.com napisał(a):
Hi Radosław,
Sorry about lost distro information, the distro we're using is Ubuntu.
We have an old copy of ceilometer container image, the ujson.so version between old and latest are both 1.35 But only latest one affected this issue.
BTW, I read the last reply on issue page. Since he said the python 3 with newer GCC is OK, I think it may caused by python version issue or GCC compiler versioning. It may become a huge architect if it really caused by compiling issue, if Ubuntu updated GCC or python.
Radosław Piliszek radoslaw.piliszek@gmail.com 於 2020年1月30日 週四 下午3:48寫道：
Hi Eddie,
the issue is that the project did *not* do a release. The latest is still 1.35 from Jan 20, *2016*... [1]
You said only Rocky source - but is this ubuntu or centos?
Also, by the looks of [2] master ceilometer is no longer affected, but monasca and mistral might still be if they call affected paths.
The project looks dead so we are fried unless we override and start using its sources from git (hacky hacky).
[1] https://pypi.org/project/ujson/#history [2] http://codesearch.openstack.org/?q=ujson&i=nope&files=&repos=
-yoctozepto
czw., 30 sty 2020 o 03:31 Eddie Yen missile0407@gmail.com napisał(a):
Hi everyone,
I'm not sure it should be bug report or not. So I email out about this issue.
In these days, I found the Rocky source deployment always failed at Ceilometer bootstrapping. Then I found it failed at ceilometer-upgrade. So I tried to looking at ceilometer-upgrade.log and the error shows it failed to import ujson.
Then I googled it and found this issue is already happened and released fixes. https://github.com/esnme/ultrajson/issues/346
But it seems like the container still using the questionable one, even today (Jan 30 UTC+8). And this not only affected to Ceilometer, but may also Gnocchi.
I think we have to patch it, but not sure about the workaround. Does anyone have good idea?
Many thanks, Eddie.