[jira] [Created] (HADOOP-15096) start-build-env.sh can create a docker image that fills up disk

["Addison Higham (JIRA)" <ji...@apache.org>]

Addison Higham created HADOOP-15096:
---------------------------------------

             Summary: start-build-env.sh can create a docker image that fills up disk
                 Key: HADOOP-15096
                 URL: https://issues.apache.org/jira/browse/HADOOP-15096
             Project: Hadoop Common
          Issue Type: Bug
          Components: build
    Affects Versions: 3.1.0
            Reporter: Addison Higham


start-build-env.sh has the potential to build an image that can fill up root disks by exploding a sparse file.

In my case, the right ingredients are:
Ubuntu 17.04
Docker 17.09.0
AUFS storage driver
userId and groupid with a high number

This happens when building the hadoop-build-${USER_ID} image, specifically in the 

{code}
RUN useradd -g ${GROUP_ID} -u ${USER_ID} -k /root -m ${USER_NAME}
{code}

command.

The reason for this:
/var/log/lastlog is a sparse file that pre-reserves based on highest seen UID and GID, in my case, those numbers are very high (above 1 billion). Locally, this result in a sparse file that reports as 443 GB. However, under docker and specifically AUFS, it appears that his file *isn't* sparse and it tries to allocate the whole file.

If you start this script and walk away to wait for it to finish, you come back to a computer with a completely full disk.

Luckily, the fix is quite easy, simply add the `-l` option to useradd which won't create those files



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org