Build container image failure is not cared
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| storlets |
New
|
Undecided
|
Unassigned | ||
Bug Description
When I ran a s2aio.sh script on clean environment, that finished with no failures, and then, tried to run some storlets apps from sample application. They all failed at emitting to launch docker container. Not yet digging deeply but looks like the reason seems that s2aio.sh script couldn't register a container image for storlets due to its permission (or user id, group id).
Another time, I tried to create swift user (e.g. useradd commeand) and then set its user_id/group_id into install/
So, I didn't get all yet but another stuff (maybe swift installation?) seems to make swift user and the user id/group id seem different from conmmon.yml-sample in s2aio.sh.
The most pain points of this are:
- Nothing failed in the setup (even if the container image registration failed)
- Storlets middleware just said "Faild to send execute command" in the log. (even if the container image is not found)
| Eran Rom (eranr) wrote | #1 |
I will try and reproduce.
Some background info:
In past versions we have tried to have the same user & group id for the user swift both inside and outside of the Docker container.
Our procedure for doing that did not always worked, and we decided to:
1. Wait until user namespaces get into the kernel (by now they are already there...), and then deal with this using user namespaces that allow to map between the container uid/gid and the hosts uid/gid
2. Until then just chmod 777 the relevant directories (I think its mainly the pipes dir).