All services like docker and k8s.io use their own namespaces
for core services, this change moves openfaas services into
the openfaas namespace instead of the default one.
The main change is that logs will look like:
journalctl -t openfaas:gateway
Instead of "default:gateway"
Function logs will remain unaffected and scheduled in the
openfaas-fn namespace.
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
Custom containers in the compose file can have a directory
mounted to store state for things like a database. This requires
a specific user since influxdb/postgresql and other containers
create folders and update permissions on start-up.
Tested with influxdb on Ubuntu with userid 1000, which failed
before the change.
Adds a grace period in the e2e tests.
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
The update flow used to delete the active function before
synchronously pulling the next and starting it. That meant
functions would always face downtime during the pull.
This changes the order to pre-pull and reduce any down time.
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
By moving to a reference instead of a value type, this patch
means the annotations are not leaked into other objects in
the result of ListFunctions. Tested on x86 with a Linux host
and I could no longer reproduce the issue in #128
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
Provide support for annotations in faasd with namespaced container
labels. Unit tested and confirmed with end to end test via faasd
deployed to multipass VM
Signed-off-by: Alex Tomic <atomic777@gmail.com>
Replace the current git checkout with a checkout of the specific
version and limit the depth to 1 for the installation.
Signed-off-by: Christopher De Vries <devries@idolstarastronomer.com>
Memory limits now work and a function will be killed with OOM
however, it will remain in a stopped state and will not
restart automatically.
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
This patch reports stopped tasks as having zero scale, which
means the gateway will send a "scale up" request, the same
way as it does for paused containers, or those which have
no task due to a reboot of the machine.
The scale up logic will now delete the stopped task and
recreate the task.
Tested with nodeinfo and figlet on a Dell XPS with
Ubuntu 16.04. The scaling logic has been re-written, but
re-tested by manually pausing and manually removing
the task of a container.
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
Logs can now be viewed with the following, adding -f to follow
the logs.
journalctl -t default:gateway
Signed-off-by: Alex Ellis (OpenFaaS Ltd) <alexellis2@gmail.com>
