Minor adjustment to how image name is set when creating a
workspace on OpenShift.
On OpenShift 3.6, it seems that using an ImageStreamTag
in a Deployment spec causes the deployment to have null status,
which causes kubernetes-client to throw an NPE.
We still need to decide if this is to be a longer term solution,
and if removing ImageStreamTags completely is a good option
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
OpenShiftConnector.inspectImage() uses pods to figure out a workspaces
docker image. Previously, this was not a problem as it would match
the imagestream tag we create earlier. However, in OpenShift 3.6 and
higher, it seems like pods created by deployments use the sha256 of
the image as an identifier, and so this fails. This means that
workspaces fail to start due to imagestream not found.
To resolve this issue, we instead get the image name through the
deployment, which does have the correct value.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Create routes in openshift with the same pattern than the urls provided by the naming strategy.
Also it will be able to handle service name described through labels on Docker image.
It means that for one image 8080 port can be tomcat, but for another image it can be named "vertx" or default value if not specified
Change-Id: Ie8e0a09d0be8f6d042f542a76c146e2bbe266fc7
Signed-off-by: Florent BENOIT <fbenoit@redhat.com>
Each pod has a dedicated folder matching the workspace id
for the project's source, it is currently using :
/projects ---> <PV>/<workspace-name>
and for the logs it's now using as well
/workspace-logs ---> <PR>/<workspace-name>-logs/
example : /workspace-logs/dev-machine-ws-agent/logs/catalina.log for the log of the dev-machine workspace agent.
Change-Id: I2b9d533ba3df2cf55857d8c30d2de5df463a5902
Signed-off-by: Florent BENOIT <fbenoit@redhat.com>
* CHE-4141 - Use Persistent Volumes Claims when creating workspaces
Signed-off-by: Snjezana Peco <snjezana.peco@redhat.com>
* Implement getContainerLogs method in OpenShiftConnector
Signed-off-by: Snjezana Peco <snjezana.peco@redhat.com>
* Implement createExec() and startExec() in OpenShiftConnector
Add implementations of createExec() and startExec(). Since OpenShift
does not separate the create and start steps, a holder class
KubernetesExecHolder is necessary, to pass information between
the call to createExec() (which just saves relevant information)
and startExec().
Additionally, adds KubernetesOutputAdapter, which parses the output
from OpenShift into LogMessages that can be handled by Che's
MessageProcessor<LogMessage> class.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Add implementation of getEvents() to avoid busy wait
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Update Dockerfile to avoid permissions issues
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* Che server and workpaces exposed on the same single TCP port (#4351)
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* Disabling usage of user account service in openshift-connector
Signed-off-by: Sun Seng David Tan <sutan@redhat.com>
* Update Docker Compose tests to fix test failure
Updating to Jackson 2.7.7 causes tests in the docker compose
plugin to fail. This is due to the fact that the tests expect
empty values in dictionaries to be parsed as the empty string,
whereas jackson 2.7.7 parses them as null (as specified by the
yaml spec).
Modifies the affected tests to explicitly use an empty string
(i.e. "") instead of an empty value.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Find an alternative to subPath in volumeMount
Signed-off-by: Snjezana Peco <snjezana.peco@redhat.com>
* Setting rwx permissions for all on /data/ in case it's not mounted
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* Add support for resource limits when running on Openshift
Add resource limits to workspace Pods when running on OpenShift.
The memory limit is normally obtained from the API request to
create the workspace, however it can be overridden via the property
`che.openshift.workspace.memory.override`. The cpu limit used is
determined by the property `che.openshift.workspace.cpu.limit`.
In both cases, the value of the property is passed directly to
OpenShift, so any valid quantity is acceptable (e.g. 150Mi,
1Gi, 1024, etc).
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Fix dockerImageConfig is null (since v1.5 of OpenShift API)
Signed-off-by: Sun Seng David Tan <sutan@redhat.com>
* Add Nullable annotation to che.docker.ip.external
The property che.docker.ip.external can be null, but
OpenShiftConnector does not include the annotation. This
prevents Che from initialising if e.g. running on docker
without the property set.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* CHE-158 Adding TLS support for Workspace routes
Signed-off-by: Ilya Buziuk <ibuziuk@redhat.com>
* Adding property to set requests for RAM
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* CHE-158 Using '-' instead of '.' for generating OpenShift route Urls
Signed-off-by: Ilya Buziuk <ibuziuk@redhat.com>
* Fixing tests after changing Url generation logic
Signed-off-by: Ilya Buziuk <ibuziuk@redhat.com>
* Redirect insecure HTTP requests to TLS endpoint
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* CHE-180: Creating and closing OpenShiftClient in every method of OpenshiftConnector
Signed-off-by: Ilya Buziuk <ibuziuk@redhat.com>
* Update route naming to make it work on OSO
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* Rework PVC management on OpenShift
- Change how subdirectories are created in pods to
use a short, terminating job instead of a full deployment
- Add OpenShiftWorkspaceFilesCleaner class to properly
notice workspace deleted events
- Add helper class to manage job pods. For creation, some
effort is made to avoid attempting to create workspaces
unnecessarily, but only exists in-memory
- Workspace deletions are batched together so that removing
workspaces directories can be done when server is idled,
avoiding unnecessary PVC mounts
- Add two new properties: che.openshift.jobs.image and
che.openshift.jobs.memorylimit, which are used by
OpenShiftPvcHelper to set up pods
Current issues:
- Since workspace directories are not deleted immediately,
attempting to re-create a workspace with the same name
will result in the previous instance's project to already
be there. This should have a minor impact.
- Memory for which workspace dirs have been created is not
persisted, resulting in potentially unnecessary jobs
- Openshift workspace files cleaner is included by overwriting
binding in WsMasterModule instead of using a provider. This
could be better, but OpenShift integration may be reaching a
point where a custom module is a better solution.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
Signed-off-by: Sun Seng David Tan <sutan@redhat.com>
* Delete ReplicaSets explicitly when shutting down a workspace
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Fix OpenShiftConnectorTest
Signed-off-by: Mario Loriedo <mloriedo@redhat.com>
* Fix route server names if unknown should start with server-.
https://issues.jboss.org/browse/CHE-230
Signed-off-by: Sun Seng David Tan <sutan@redhat.com>
* Add property to control manual workspace dir creation in OpenShift
Add property 'che.openshift.precreate.workspace.dirs'. If property is
true, OpenShiftConnector will run a pod before launching workspaces
to create a subpath in the workspace's persistent volume with correct
permissions. If the property is false, this step is skipped.
This is necessary as in older versions of OpenShift/Kubernetes, subpaths
created as part of a volume mount are created with root permissions, and
so cannot be modified by workspace pods. More recent versions fix this,
creating subpath volumes with correct permissions, making the step above
unnecessary.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* CHE-102 - Idle detection of che-server and workspaces
Signed-off-by: Snjezana Peco <snjezana.peco@redhat.com>
* Add and modify tests for OpenShift helper classes
Add tests for the untested classes in openshift.client.kuberentes,
and update existing tests where necessary.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
* Recent changes required access to `/` which is impossible under OS
Signed-off-by: David Festal <dfestal@redhat.com>
* adapt che-server entrypoint.sh to environments without write permissions in '/' (#5344)
* adapt che-server entrypoint.sh to environments without write permissions in '/'
* CHE-280: Adding container's state info to the 'inspectContainer' API
Signed-off-by: Ilya Buziuk <ibuziuk@redhat.com>
* Factorize code of `ServerEvaluationStrategy` classes, to use the Custom strategy as the basis of other strategies (#5366)
* Pull-up the local docker port management (use exposed ports)
Signed-off-by: David Festal <dfestal@redhat.com>
* Make all the strategies extend `CustomEvaluationStrategy`
Signed-off-by: David Festal <dfestal@redhat.com>
* Add a `workspaceIdWithoutPrefix` macro and use it for `single-port`
This macro is based on the `workspaceId` macro, but without the
`workspace` prefix.
Signed-off-by: David Festal <dfestal@redhat.com>
* Add the `isDevMachine` to allow conditions in the ST template.
This is required to allow the `single-port` strategy to have a different
url according to the type of machine. (see the work done for CHE-175 :
Support multi-container workspaces on OpenShift)
Signed-off-by: David Festal <dfestal@redhat.com>
* Small fixes after comments from @fbenoit
Signed-off-by: David Festal <dfestal@redhat.com>
* Fix unnecessary space pointed out by @sunix
Signed-off-by: David Festal <dfestal@redhat.com>
* Remove unnecessary `else` as suggested by @sunix
Signed-off-by: David Festal <dfestal@redhat.com>
* Keep the method signatures compatible with the `condenvy` strategy
Signed-off-by: David Festal <dfestal@redhat.com>
* Align names of parameters of constructors (requested by @garagatyi)
Signed-off-by: David Festal <dfestal@redhat.com>
* Add a default implementation to avoid breaking the Codenvy build
Signed-off-by: David Festal <dfestal@redhat.com>
* Also rename the attributes
Signed-off-by: David Festal <dfestal@redhat.com>
* Use a constant for the `workspace` prefix string
Signed-off-by: David Festal <dfestal@redhat.com>
* Fix formatting as requested by @sunix
Signed-off-by: David Festal <dfestal@redhat.com>
* Use a constant for the `isDevMachine` macro name
Signed-off-by: David Festal <dfestal@redhat.com>
* Add unit tests for `workspaceIdWithoutPrefixè and `isDevMachine` macros
Signed-off-by: David Festal <dfestal@redhat.com>
* Another requested formatting fix
Signed-off-by: David Festal <dfestal@redhat.com>
* Make new tests clearer
Signed-off-by: David Festal <dfestal@redhat.com>
* yet another formatting request
Signed-off-by: David Festal <dfestal@redhat.com>
* Respect the original order of imports
Signed-off-by: David Festal <dfestal@redhat.com>
* remove unnecessary `toString()`
Signed-off-by: David Festal <dfestal@redhat.com>
* use a lowercase `S` in the `server-` prefix
Signed-off-by: David Festal <dfestal@redhat.com>
* Multi-container workspace Support (#5110)
* Fix 2 NPE that prevented using *non-dev* additional machines
In the context of https://issues.jboss.org/browse/CHE-175
Signed-off-by: David Festal <dfestal@redhat.com>
* Name openshift resources based on the machine name for non-dev machines
This fixes https://issues.jboss.org/browse/CHE-259
and https://issues.jboss.org/browse/CHE-258
Signed-off-by: David Festal <dfestal@redhat.com>
* Fix failing Traeffik tests...
... by:
- adding the new `CHE_IS_DEV_MACHINE` env variable in tests
- pulling up all the `CustomServerEvaluationStrategy` features in an
abstract `BaseServerEvaluationStrategy` (which all other Che strategies
extend) and have the `CustomServerEvaluationStrategy` class simply
extend this `BaseServerEvaluationStrategy`.
Signed-off-by: David Festal <dfestal@redhat.com>
* Fix tests in the LocalDockerEvaluationStrategy...
... by correctly using the boolean attribute to manage the new use-case
introduced by @fbenoit in master.
Signed-off-by: David Festal <dfestal@redhat.com>
* Replace OSIO-specific `single-port` strategy by `docker-local-custom`
This fixesredhat-developer/rh-che#113
Signed-off-by: David Festal <dfestal@redhat.com>
Prevents issue where it is possible to start workspace while
pod is terminating. When this occurs, Che creates a new pod but
workspace start fails. This pod must be removed manually before
that workspace can be launched again.
This is done by a) waiting for pod to terminate before returning
from removeContainer, and b) removing created deployment and
service when an exception occurs during createContainer.
Additionally, increases maximum wait time for retreiving
OpenShift image stream metadata
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
Adds commit() and removeImage() implementations to
OpenShiftConnector. This requires some refactoring of existing
methods (mostly refactoring repeated processes into their own
methods -- e.g. creating an ImageStreamTag and getting image
info from a tag).
Additionally, refactors inspectContainer() method to remove
a call to DockerConnector, instead obtaining the same information
from what's available through the OpenShift API. This fixes an
issue where the IP address of a workspace was unavailable from
the ContainerInfo returned by DockerConnector.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
Implement OpenShiftConnector pull, tag, and inspectImage through
ImageStreams. Makes changes to how containers are created.
Currently snapshots do not work, as commit and push methods are not
implemented. Additionally, pulling a stack from a private repository
is not supported.
Signed-off-by: Angel Misevski <amisevsk@redhat.com>
Roman Iuvshyn