You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@singa.apache.org by "Ngin Yun Chuan (JIRA)" <ji...@apache.org> on 2019/03/04 08:28:00 UTC
[jira] [Comment Edited] (SINGA-430) Rafiki--Creat inference error.

    [ https://issues.apache.org/jira/browse/SINGA-430?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16783101#comment-16783101 ] 

Ngin Yun Chuan edited comment on SINGA-430 at 3/4/19 8:27 AM:
--------------------------------------------------------------

Hi Liu Hui,

Whenever a new inference job is created, Rafiki selects a free port to use for the job's predictor, starting from port 30000, and it determines which ports are free based on its database of inference jobs. If there was an existing Docker container/service using a port not captured in that instance of Rafiki, there'll be that error. 

Currently, this can be avoided by killing all applications using ports >= 30000 on the deployed machine.

In Rafiki's code, maybe we should better handle the error to try successive ports, if creation of a predictor at a port fails. Will be adding this to our list of TODOs.

Thanks for the report!



was (Author: nginyc):
Hi Liu Hui,

Whenever a new inference job is created, Rafiki selects a free port to use for the job's predictor, starting from port 30000, and it determines which ports are free based on its database of inference jobs. If there was an existing Docker container/service using a port not captured in that instance of Rafiki, there'll be that error. 

Currently, this can be avoided by killing all applications using ports > 30000 on the deployed machine.

In Rafiki's code, maybe we should better handle the error to try successive ports, if creation of a predictor at a port fails. Will be adding this to our list of TODOs.

Thanks!


> Rafiki--Creat inference error.
> ------------------------------
>
>                 Key: SINGA-430
>                 URL: https://issues.apache.org/jira/browse/SINGA-430
>             Project: Singa
>          Issue Type: Bug
>            Reporter: Liu Hui
>            Priority: Major
>         Attachments: err.log
>
>
> When I ran a task of image identification on rafiki which showed in Rafiki's User Guide.
> I have created and trained  model successfully.
> But a error occurred when I try to create an inference.The error log shows in attachment.
> The hint is "port '30000' is already in use",I think that the problem happened when docker swarm try to create new docker for inference.
>  
> Issue  can be tested on the machine 221.224.36.165:/home/liuhui/rafiki.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)