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.
>
> https://pastebin.com/nGqsM0uf
>
> 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.