I have the following setup:. Perl service running in a container and writing logs out to STDERR. to ship those logs out to a remote server for archiving in a 600 MB RAM machine. I also truncate the logs periodically at: /var/lib/docker/containers/CID/CID-json.log as suggested to avoid 100% disk scenarios. Problem Docker daemon starts of with low memory usage, 1% initially and slowly increases to 40% after 2 days of running the container. Reference Docker daemon memory leak has been talked about in.
- Over 2 years Memory leak when using Parallels + volumes over 2 years file could not mount into boot2docker over 2 years Cannot connect to docker-machine with boot2docker 1.11.0.
- Assume that /tmp/link is a soft link to /home/link, if we use /tmp/link/docker as a graph path, and /tmp/link/docker is not exist, docker can not get the real path.
Docker logs webserver - inspect the logs; Docker logs -f webserver -- follows the logs on the container; Docker attach webserver - attaches to the running container in detached mode but the disadvantage is if you attached and you exit the container shell then container will exit too.
Com.docker.osxfs Memory Leak Issue 1815 Docker/for-machine
But both of them are closed now saying merged at a commit. Am running the latest major version of docker (Docker version 1.4.0, build 4595d4f), but still face a monotonically increasing memory usage issue.
EDIT: I did this experiment: Just run a bash process in the container, print out a lot of lines to STDERR, docker daemon process's memory usage accelerates very quickly Does docker do some and doesnt release memory even if underlying log file (/var/lib/docker/containers/CID/CID-json.log) is cleared? There's apparently to clear the logs. Will solve this issue for long running tasks? I dont know why docker daemon's memory usage keeps increasing. How do I debug this issue? This may not be what you are looking for, but I usually run a cron job to restart my containers after a certain amount of time everyday. This ensures that the container has enough RAM all the time, and also I generally restrict the maximum ram usage by the container while creating them.
Containers take only few seconds to restart and serve data and if you are not running a High Availability service and can afford a few seconds downtime, consider restarting the container (assuming that you dont have persistent volumes). However, if you do find a solution to your problem, do let us know.
Signed-off-by: Akihiro Suda 
RUN TestTarUntarWithXattr - PASS: TestTarUntarWithXattr (0.01s) PASS coverage: 30.7% of statements ok github.com/docker/docker/pkg/archive 0.018s coverage: 30.7% of statements Tested on Ubuntu 18.04.1, kernel 4.15.0-39-generic #42-Ubuntu - Description for the changelog
What I did Add config flags to allow “linuxkit run qemu” to mount host USB devices. How I did it The “usb” flag enables the qemu USB host controller, and “devices” multi-flag to specify each device added. This doesn’t work for “containerized” qemu runs since that requires passing devices to docker as well. How to verify it linuxkit run qemu -usb -device usb-ehci,id=ehci -device usb-host,vendorid=0x0001,productid=0x0001 - Description for the changelog Add USB host devices to linuxkit qemu run. Assume that /tmp/link is a soft link to /home/link, if we use /tmp/link/docker as a graph path, and /tmp/link/docker is not exist, docker can not get the real path.
Addresses: - How I did it I started off by looking at existing packages for a reference model of how to structure a package in linuxkit. I also read the doc. How to verify it You can build it locally to test it out using the following command: linuxkit pkg build -org= -disable-content-trust - Description for the changelog This commit adds bpftrace to linuxkit packages and addresses the issue: - A picture of a cute animal (not mandatory but encouraged) Signed-off-by: Simarpreet Singh.
Adds a set/map with the services to be ignored by the render command. All other usages of render.Render just pass ‘nil’. Note that it’s a WIP PR. Maybe we could use a better name for the flag and use the concept of filtering for other concepts.
Closes #396 - What I did Added the capability to remove services from the render by specifying them in the command line. How I did it Added a new flag in the “render” command called “disable” - How to verify it Given an already existing application package that contains 3 services called “service1”, “service2” and “service3” run: docker-app render -disable service1 -disable service2 Only “service3” should be rendered - Description for the changelog Add a flag to the render command and pass the service names collected through the render.Render to be filtered afterwards. Signed-off-by: Philipp Schmied 
This is now being called at the start of the functions implementing the cp, export and save commands. For cp operations to a container, the fix only uses the helper method mentioned above, so it’s possible to unify this check for all three operations. The helper method also determines whether the target path points to a device or any other invalid/irregular file to display an error message accordingly. Also, I’ve added test cases for all commands.
How to verify it. Clone the cli master branch and compile a static docker binary. Pull an arbitrary docker image: docker pull ubuntu:latest. Invoke docker save ubuntu:latest -o /dev/random. Check the output of stat /dev/random indicating /dev/random is now a regular file.
Perform the steps above using my PR and see that this operation is prevented, yielding failed to save image: invalid output path: '/dev/random' must be a directory or a regular file: Got a device. Description for the changelog
As for the shell, it’s nice to be able to use bash instead of the more bare-bones ash if preferred. Finally, being able to name the container can come in handy for easier scripting on the host. This patch won’t change anything for anyone who doesn’t set these env variables in their environment.
How to verify it The output of: make -f docker.Makefile -n shell should be unchanged compared to what it was before this patch; but: DOCKERCLIMOUNTS='-v test:/test' make -f docker.Makefile -n shell should replace the -v flag from the command above with just -v test:/test, and finally: DOCKERCLIMOUNTS=' ' make -f docker.Makefile -n shell should be devoid of any -v flags (besides the socket one). All the make targets should be unchanged when not setting any of the new env vars. A picture of a cute animal (not mandatory but encouraged).
Signed-off-by: Andre Fernandes Proposal: DOCKERHOSTSSHOPTIONS environment variable. What I did The new “ssh://” protocol for DOCKERHOST is great, but sometimes we need to use different ssh options than the default. A common scenario if the need to point to a project-specific ssh config file instead of the default one at “/.ssh/config”. How I did it An environment variable DOCKERHOSTSSHOPTIONS that can hold options for the ssh command docker uses when DOCKERHOST contains the “ssh://” prefix. How to verify it export DOCKERHOSTSSHOPTIONS=“-F /my/own/ssh.cfg” export DOCKERHOST=“ssh://somehost” docker version. Description for the changelog DOCKERHOSTSSHOPTIONS to change ssh options when using “ssh://” protocol. Performing syscall lchown in most unix systems will clear setuid/setgid bit from file Fixes: #36239 - What I did Fix bug with missing setuid/setgid bits on files in resulting image after docker build command - How I did it In function fixPermissions we perform copy of source file permissions after issuing lchown syscall.
How to verify it I have used the following script to verify it in my build of docker: ```!/bin/bash cd /opt touch 1.bin chmod 700 1.bin chmod u+xs 1.bin echo -e ‘FROM busybox:latest nADD 1.bin /1.bin’ Dockerfile docker build -t test. ``` - Description for the changelog Fix issue with setuid/setgid bits on files when copy files during docker build process.