diff options
Diffstat (limited to 'k8s/k8splugin/tasks.py')
-rw-r--r-- | k8s/k8splugin/tasks.py | 93 |
1 files changed, 59 insertions, 34 deletions
diff --git a/k8s/k8splugin/tasks.py b/k8s/k8splugin/tasks.py index 50087fb..4205122 100644 --- a/k8s/k8splugin/tasks.py +++ b/k8s/k8splugin/tasks.py @@ -30,11 +30,11 @@ from cloudify.exceptions import NonRecoverableError, RecoverableError from onap_dcae_dcaepolicy_lib import Policies from k8splugin import discovery as dis from k8splugin.decorators import monkeypatch_loggers, wrap_error_handling_start, \ - merge_inputs_for_start, merge_inputs_for_create + merge_inputs_for_start, merge_inputs_for_create, wrap_error_handling_update from k8splugin.exceptions import DockerPluginDeploymentError from k8splugin import utils from configure import configure -from k8sclient import k8sclient +import k8sclient # Get configuration plugin_conf = configure.configure() @@ -245,7 +245,7 @@ def _verify_container(service_component_name, max_wait): will be raised. """ num_attempts = 1 - + while True: if k8sclient.is_available(DCAE_NAMESPACE, service_component_name): return True @@ -256,7 +256,7 @@ def _verify_container(service_component_name, max_wait): raise DockerPluginDeploymentError("Container never became healthy") time.sleep(1) - + return True def _create_and_start_container(container_name, image, **kwargs): @@ -266,7 +266,7 @@ def _create_and_start_container(container_name, image, **kwargs): We're not exposing k8s to the component developer and the blueprint author. This is a conscious choice. We want to use k8s in a controlled, consistent way, and we want to hide the details from the component developer and the blueprint author.) - + kwargs may have: - volumes: array of volume objects, where a volume object is: {"host":{"path": "/path/on/host"}, "container":{"bind":"/path/on/container","mode":"rw_or_ro"} @@ -287,21 +287,21 @@ def _create_and_start_container(container_name, image, **kwargs): ctx.logger.info("Deploying {}, image: {}, env: {}, kwargs: {}".format(container_name, image, env, kwargs)) ctx.logger.info("Passing k8sconfig: {}".format(plugin_conf)) replicas = kwargs.get("replicas", 1) - _,dep = k8sclient.deploy(DCAE_NAMESPACE, - container_name, + _,dep = k8sclient.deploy(DCAE_NAMESPACE, + container_name, image, - replicas = replicas, + replicas = replicas, always_pull=kwargs.get("always_pull_image", False), k8sconfig=plugin_conf, - volumes=kwargs.get("volumes",[]), + volumes=kwargs.get("volumes",[]), ports=kwargs.get("ports",[]), - msb_list=kwargs.get("msb_list"), + msb_list=kwargs.get("msb_list"), env = env, labels = kwargs.get("labels", {}), log_info=kwargs.get("log_info"), readiness=kwargs.get("readiness")) - # Capture the result of deployment for future use + # Capture the result of deployment for future use ctx.instance.runtime_properties["k8s_deployment"] = dep ctx.instance.runtime_properties["replicas"] = replicas ctx.logger.info ("Deployment complete: {0}".format(dep)) @@ -337,11 +337,11 @@ def _enhance_docker_params(**kwargs): Set up Docker environment variables and readiness check info and inject into kwargs. ''' - + # Get info for setting up readiness probe, if present docker_config = kwargs.get("docker_config", {}) if "healthcheck" in docker_config: - kwargs["readiness"] = docker_config["healthcheck"] + kwargs["readiness"] = docker_config["healthcheck"] envs = kwargs.get("envs", {}) @@ -364,7 +364,7 @@ def _enhance_docker_params(**kwargs): # lists together with no deduping. kwargs = combine_params("ports", docker_config, kwargs) kwargs = combine_params("volumes", docker_config, kwargs) - + return kwargs @@ -375,15 +375,15 @@ def _create_and_start_component(**kwargs): # Need to be picky and manually select out pieces because just using kwargs # which contains everything confused the execution of # _create_and_start_container because duplicate variables exist - sub_kwargs = { + sub_kwargs = { "volumes": kwargs.get("volumes", []), "ports": kwargs.get("ports", None), - "envs": kwargs.get("envs", {}), + "envs": kwargs.get("envs", {}), "log_info": kwargs.get("log_info", {}), "labels": kwargs.get("labels", {}), "readiness": kwargs.get("readiness",{})} _create_and_start_container(service_component_name, image, **sub_kwargs) - + # TODO: Use regular logging here ctx.logger.info("Container started: {0}".format(service_component_name)) @@ -392,12 +392,7 @@ def _create_and_start_component(**kwargs): def _verify_component(**kwargs): """Verify component (container) is healthy""" service_component_name = kwargs[SERVICE_COMPONENT_NAME] - # TODO: "Consul doesn't make its first health check immediately upon registration. - # Instead it waits for the health check interval to pass." - # Possible enhancement is to read the interval (and possibly the timeout) from - # docker_config and multiply that by a number to come up with a more suitable - # max_wait. - + max_wait = kwargs.get("max_wait", 300) # Verify that the container is healthy @@ -407,7 +402,7 @@ def _verify_component(**kwargs): # TODO: Use regular logging here ctx.logger.info("Container is healthy: {0}".format(service_component_name)) - + return kwargs def _done_for_start(**kwargs): @@ -499,7 +494,7 @@ def create_and_start_container_for_platforms(**kwargs): image = ctx.node.properties["image"] docker_config = ctx.node.properties.get("docker_config", {}) if "healthcheck" in docker_config: - kwargs["readiness"] = docker_config["healthcheck"] + kwargs["readiness"] = docker_config["healthcheck"] if "dns_name" in ctx.node.properties: service_component_name = ctx.node.properties["dns_name"] else: @@ -564,7 +559,7 @@ def create_and_start_container(**kwargs): image = ctx.node.properties["image"] _create_and_start_container(service_component_name, image,**kwargs) - + ctx.logger.info("Component deployed: {0}".format(service_component_name)) @@ -580,6 +575,7 @@ def stop_and_remove_container(**kwargs): ctx.logger.error("Unexpected error while stopping container: {0}" .format(str(e))) +@wrap_error_handling_update @monkeypatch_loggers @operation def scale(replicas, **kwargs): @@ -587,15 +583,44 @@ def scale(replicas, **kwargs): if replicas > 0: current_replicas = ctx.instance.runtime_properties["replicas"] ctx.logger.info("Scaling from {0} to {1}".format(current_replicas, replicas)) - try: - deployment_description = ctx.instance.runtime_properties["k8s_deployment"] - k8sclient.scale(deployment_description, replicas) - ctx.instance.runtime_properties["replicas"] = replicas - except Exception as e: - ctx.logger.error ("Unexpected error while scaling {0}".format(str(e))) + deployment_description = ctx.instance.runtime_properties["k8s_deployment"] + k8sclient.scale(deployment_description, replicas) + ctx.instance.runtime_properties["replicas"] = replicas + + # Verify that the scaling took place as expected + max_wait = kwargs.get("max_wait", 300) + service_component_name = ctx.instance.runtime_properties["service_component_name"] + if _verify_container(service_component_name, max_wait): + ctx.logger.info("Scaling complete : {0} from {1} to {2} instance(s)".format(service_component_name, current_replicas, replicas)) + else: ctx.logger.info("Ignoring request to scale to zero replicas") - + +@wrap_error_handling_update +@monkeypatch_loggers +@operation +def update_image(image, **kwargs): + if image: + current_image = ctx.instance.runtime_properties["image"] + ctx.logger.info("Updating application container image from {0} to {1}".format(current_image, image)) + deployment_description = ctx.instance.runtime_properties["k8s_deployment"] + k8sclient.upgrade(deployment_description, image) + ctx.instance.runtime_properties["image"] = image + + # Verify that the update took place as expected + max_wait = kwargs.get("max_wait", 300) + service_component_name = ctx.instance.runtime_properties["service_component_name"] + if _verify_container(service_component_name, max_wait): + ctx.logger.info("Update complete : {0} from {1} to {2} instance(s)".format(service_component_name, current_image, image)) + + else: + ctx.logger.info("Ignoring update_image request with unusable image '{0}'".format(str(image))) + +#TODO: implement rollback operation when kubernetes python client fix is available. +# (See comments in k8sclient.py.) +# In the meantime, it's possible to undo an update_image operation by doing a second +# update_image that specifies the older image. + @monkeypatch_loggers @Policies.cleanup_policies_on_node @operation @@ -628,7 +653,7 @@ def _notify_container(**kwargs): pods could be changing. We can query to get all the pods, but there's no guarantee the list won't change while we're trying to execute the script. - + In ONAP R2, all of the policy-driven components rely on polling. """ """ |