- Mailing Lists
- Contributors
- Re: Warning: Using "numpy" library you will have weird errors in production even if you are not using it
Archives
- By thread 1419
-
By date
- August 2019 59
- September 2019 118
- October 2019 165
- November 2019 97
- December 2019 35
- January 2020 58
- February 2020 204
- March 2020 121
- April 2020 172
- May 2020 50
- June 2020 158
- July 2020 85
- August 2020 94
- September 2020 193
- October 2020 277
- November 2020 100
- December 2020 159
- January 2021 38
- February 2021 87
- March 2021 146
- April 2021 73
- May 2021 90
- June 2021 86
- July 2021 123
- August 2021 50
- September 2021 68
- October 2021 66
- November 2021 74
- December 2021 75
- January 2022 98
- February 2022 77
- March 2022 68
- April 2022 31
- May 2022 59
- June 2022 87
- July 2022 141
- August 2022 38
- September 2022 73
- October 2022 152
- November 2022 39
- December 2022 50
- January 2023 93
- February 2023 49
- March 2023 106
- April 2023 47
- May 2023 69
- June 2023 92
- July 2023 64
- August 2023 103
- September 2023 91
- October 2023 101
- November 2023 94
- December 2023 46
- January 2024 75
- February 2024 79
- March 2024 104
- April 2024 63
- May 2024 40
- June 2024 160
- July 2024 80
- August 2024 70
- September 2024 62
- October 2024 121
- November 2024 117
- December 2024 89
- January 2025 59
- February 2025 104
- March 2025 96
- April 2025 107
- May 2025 52
- June 2025 72
- July 2025 60
- August 2025 81
- September 2025 124
- October 2025 63
- November 2025 22
Contributors
Re: Warning: Using "numpy" library you will have weird errors in
production even if you are not using it
Thanks Moises for the heads up!
I faced a similar case.
A (non-OCA) module was using openpyxl,
which loads numpy when available,
which loads libopenblasp
OpenBLAS creates a number of threads equal to the number of core threads available: 56 in my case (production server),
so it quickly reached limit_memory_hard
and the process was killed (SIGSEGV)
Forcing OPENBLAS_NUM_THREADS=1 fixed the issue.
Regards,
On Mon, Nov 2, 2020 at 3:12 PM Stéphane Bidoul <stephane.bidoul@acsone.eu> wrote:
Hi Moises,It's indeed unfortunate that you have had such issues. It's also very surprising that websocket-client attempts to import numpy :)But I don't think blacklisting numpy is appropriate. It is very widely used outside OCA and can be useful at times in OCA too.-sbiOn Mon, Nov 2, 2020 at 9:01 AM Pedro M. Baeza (Tecnativa) <pedro.baeza@tecnativa.com> wrote:Thanks for the warning.I don't use any OCA module using numpy AFAIK, but there are some out there indeed. Not sure if blacklisting is the best thing to do, as changing such modules to adapt to other things may not be easy, but developers that used it can tell.Regards._______________________________________________
Mailing-List: https://odoo-community.org/groups/contributors-15
Post to: mailto:contributors@odoo-community.org
Unsubscribe: https://odoo-community.org/groups?unsubscribe
_______________________________________________
Mailing-List: https://odoo-community.org/groups/contributors-15
Post to: mailto:contributors@odoo-community.org
Unsubscribe: https://odoo-community.org/groups?unsubscribe
by Nils Hamerlinck - 08:06 - 30 Jan 2021
Reference
-
Warning: Using "numpy" library you will have weird errors in production even if you are not using it
Hello OCA Contributors,numpy is a powerful math libraryBut there are weird issues that could be a nightmare in production to found them.Imagine that you start your odoo-bin and during starting you just see the following message:2020-10-29 15:13:15,053 194 INFO openerp_test odoo.modules.loading: updating modules list
Segmentation fault (core dumped)You can run gdb to find more details but it shows:[New Thread 0x7fb752885700 (LWP 1630)]
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fb752885700 (LWP 1630)]
0x0000000000000000 in ?? ()Really, WT* is that?Where is the issue man?After 3 days of my weekend without sleep and running line by line I found that the impostor was: numpyI just uninstalled and now all is working well.The weird part is that this is a corner case issue and it is not reproduced easily.Virtualization environment with a particular kind of processor.In our case it was reproduced only and only in production environment (no staging ones)Another issue was that all processors were used 100%We just uninstalled numpy and now all is working well.The issue is only using this line: "import numpy"Even if you are not using that odoo module it line is executed and all your environment will be broken.More info about the issues:- Vx/MQT#315 (Big debug history)- numpy#17674 (more recent)- numpy#13059 (guilty I didn't sleep for 3 days)The following libraries are using numpy as dependency:- pandas- bokeh- altair- websocketSo the following OCA projects depends of numpy:- OCA/webIn fact, numpy is deprecating financial operations, we can use the following library:IMHO we should add to "numpy" in the blacklist of libraries in the OCAAnd look for other alternatives.For pandas, I know it could increase the performance to process too much recordsbut in Odoo we have PostgreSQL and maybe we could find similar results.Maybe a pipeline warning for our CI could be helpful if this library is installed.What do you think?--Moisés López CalderónMobile: (+52) 477-752-22-30Twitter: @moylop260Twitter: @vauxoo
by Moisés López Calderón - 06:35 - 30 Oct 2020