- added retainPods flag to take in the number of specialized pods to retain
- add retainPods in both the create function and update function command
- modify crd keys to be typed instead of string
- keep track of function generation in case of update function operation
- add delete handler function to make sure specialized pods are deleted in case function is deleted
---------
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Signed-off-by: Pranoy Kundu <pranoy1998k@gmail.com>
Co-authored-by: Pranoy Kundu <pranoy1998k@gmail.com>
* add functionality to wait for specialization by keeping track of incoming requests
* format executor package
* fix required capacity to specialise new pod condition
* move handling concurrency logic into pool cache from executor
* remove unused methods and structs
* implement queue in to store the svc wait
* create a queue struct and its methods to handle concurrent inputs
* use newly created queue to store waiting for svc requests
* add waiting requests in queue and use them when a svc is ready
* set function to request in queue if the context is still alive
* remove concurrency approach to set svc for waiting requests
* update the active requests whenever requests from pool are assigned a svc
* add doc to define why the conditions exist
* remove unwanted params in strcut and clean up code
* set error while getting svc value if sum of specialization in progress and specialized is only more than concurrency limit
* remove duplicate functions and unnecessary values in struct
* close svc channel on set value and create constants for default concurrency and rpp
* get next value in queue in case context is timed out for fetched value
* remove specializationInProgress counter from pool cache
* return in case the queue is empty wihle setting func to svc
* test getSvcVaue and setSvcValue in poolcache
* add unit tests for GetConcurrent and GetRequestsPerPod methods
* reorder imports
* add fuzzy testing for getSVCValue and setSVCValue in poolcache
* restructure go mod file and update pool cache test cases
* Add tests and bug fixes
* refactor code and add test cases
* add svcWaiting check while setting svc value
---------
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Co-authored-by: Sanket Sudake <sanketsudake@gmail.com>
* service account changes for fission-builder
* moved code into executor
* changed default timing to 30 min
* code refactor and add new roles for executor
* create role if serviceAccountCheck is enabled
* Use informerfactory across executor
* Run function informer for poolpodcontroller if istio enabled
* Use same namespace for secret as keda mqtriggers
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
* Remove cluster role references
* Convert secret/package getter cluster roles to role
* Remove all cluster role binding permissions
* Remove unwanted permission
* Try removing all RBAC-related code
* Remove additional constants
* Add functionality for service account checks
* Restrict permission across services
* Remove package informer factory from the executor
* Revert service account check code for now
* Skip adding roles in place of cluster roles
* Remove additional permission from the router
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Consider specific namespaces mentioned by the user in building informers in the executor
- Confimaps
- Secrets
- Deployments
- Services
- Pods
- Replicasets
Currently, we create Fission resources in the default namespace, function-related resources are created in the fission-function namespace, whereas builder resources are created in the fission-builder namespace. This causes confusion for a lot of users.
In this fix, we allow the user to set the function and builder namespace empty so that function and builder resources are created in the same namespace as the function resource always.
If the user desires older behaviour they can functionNamespace and builderNamespace the same previous before the upgrade.
* use default namespace for fission function and builder
* support for existing fission namespaces
* Replace builder and function namespace with template
* Fix namespace creation template
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
We allow functions to refer to configmap/secrets. We are monitoring all namespaces for config maps and secrets and also allow cross-namespace references.
This fix monitors configmaps/secret updates in specific namespaces. Also, we ignore cross-namespace references for configmap/secret updates.
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
The users can now set the pod spec for builder and fn pods via helm chart.
Currently we have set some default securitycontext for the pods. Before there were no permissions set and the user would by default enter root when kubectl exec into pod. Now the permissions have been set and the user will not be able to access root directory in poolmgr and newdeploy pods.
Using interface makes it easy to create a fake client and unit test
a specific portion of the code. We should be able to more write unit
test and increase coverage of code with this change.
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
* Defining httpserver package to capture httpserver shutdown and
introduces uniform running of http server across codebase.
* Add unit tests for httpserver
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
- Enabled metrics in storagesvc, buildermgr and controller.
- Added a middleware in storagesvc, router, executor and controller to monitor total number of http requests, each request's duration and number of requests that are currently being served. These requests can be filtered on their path, method or statuscode.
- Removed functionCallDuration and functionCallResponseSize metrics from router.
- Removed funcAliveSummary, funcIsAlive, funcReapTime and idleTime metrics.
- Replaced function calls for collecting metrics to direct metric calls.
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Co-authored-by: Sanket Sudake <sanketsudake@gmail.com>
- Currently, fission components don't handle shutdown signals.
So we don't get any to do the required cleanup before the fission process
exits. Adding signal capture process with cancelling context so
that all dependent processes stop working when the process gets term
signal.
- Set log level to error in otel shutdown function
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
* Capture important open telemetry events with span in fission
* Add context to missing HTTP calls
* Add Trace ID in logs
* capture trace id in the proxy handler function
* Always registry tracer to get traceID
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Signed-off-by: Sanket Sudake sanketsudake@gmail.com
- Use informers and listers in executors
- Passing context properly in pool manager executor
- Environment updates in the pool manager would not cause updates in the deployment
- Environment update minimizing downtime
- Use replicaset controller and environment delete triggers to cleanup specialized pods
Generally using typed informers is more standard practise
than using SharedIndexInformer(SII) implicity. SII also lack
listers provided by informer factory and few other high level
abstractions.
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Currently most of informers used in Fission are using resync
period of 30sec which is quite low and causes lot of disturbances
in overall system. Since informers already have watchers which would
be delivering events to the handlers. If any handlers require lower
sync period they should register handler with `AddEventHandlersWithResyncPeriod`
instead of we generaling informer.
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
* add opentracing section and otelCollectorEndpoint
* initialize OTLP exporter
* pkg/controller: changes for context propagation
* pkg/executor: changes for context propagation
* pkg/fetcher: changes for context propagation
* pkg/router: changes for context propagation
* pkg/storagesvc: changes for context propagation
* set no default value for otel collector endpoint
* update readme and add notes to charts
* move common code to pkg/utils/otel
* adding fn and env as attributes
* don't use otelhttp transport for websocket
* URL ignore with common filter UrlsToIgnore
Note: The web socket example does not work when using OTEL HTTP. Here is an issue related to that on open-telemetry/opentelemetry-js-contrib.
Signed-off-by: Gaurav Gahlot <gauravgahlot0107@gmail.com>
Co-authored-by: Sanket Sudake <sanketsudake@gmail.com>
PR adds a new executortype which supports running containers as functions. New CLI under functions is added to create container as functions.
Co-authored-by: Harsh Thakur <harshthakur9030@gmail.com>
Co-authored-by: Sanket Sudake <sanketsudake@gmail.com>
* Add a SharedIndexInformer for services and deployments to NewDeploy executor.
This brings the NewDeploy executor behaviour into sync with GenericPoolManager behaviour
by caching Kubernetes services and deployments used in per-request function validation.
* Create informers in executer by executor label
Signed-off-by: Sanket Sudake <sanketsudake@gmail.com>
Co-authored-by: James Sinclair <irgeek@btc.com.au>
This feature enables routing more than one request to a pod at the same time. This is the first draft of the work and might involve more optimizations later.
Concurrency in the pool manager allows specializing pods concurrently based on a specified limit.
Co-authored-by: Vishal <vishal-biyani@users.noreply.github.com>
Previously, once the executor is deleted for reasons (like upgrade or cluster scale-in),
the new executor deletes all existing resources created by the old executor and creates
new one. This mechanism becomes a problem when there are requests connecting to the
existing pods. Also in the worst case, the cluster may not have enough resources to create
new pods and cause service downtime.
This PR let each executor type adopts existing resources before starting the executor
API services, and so the alive connections won't experience failure. However, the requests
send to the function that doesn't have alive function pods will still fail due to the
executor is in bootstrapping.
The pool manager keeps terminating function pod periodically even there are
traffic to the function. The root cause is that executor, poolmgr, newdeploy
manage their own functionServiceCache separately. And when router taps a
function, executor updates the access time of the function service entry in its
own cache without notifying executor types to do the update as well. Hence,
the access time of function service entry in poolmanager cache never gets updated.
Due to the access time never gets updated, the idle pod reaper in poolmanager
then thinks the function pod is in idle state and recycle it.
This PR removes the cache in executor itself, and when router tries to tap a function,
executor will call executor type to tap the function and update access time.
When a function with executor type newdeploy got created, Newdeploy
is expected to create deployment/HPA/service for it and insert serviceEntry
to the cache for later use. Once clients call the function, newdeploy returns
the serviceEntry to the router.
However, the log shows that the newdeploy was unable to find the entry and
prints "Resource not found - key 'xxx' not found". The root cause is that the
informer controller instead of processing items in parallel, it dispatches XXFunc
to process items one by one. So if there is any problem during the creation of the
kubernetes resource, it takes a longer time to process the next item and hence
the serviceEntry was not inserted before clients call the function.
This PR lets the newdeploy to process items in extra goroutines instead of blocking
the process loop. It's a workaround to solve the problem above, we should consider
using workqueue to solve it in the future.
If a user deploys fission in the namespace which is different from the one in the single YAML file generated by helm, fission components won't be able to talk to each other due to the wrong namespace appends after the service address.
This PR adds --namespace when generating the YAML file to prevent the mismatch problem.
* Config and secret change to invoke function update
* Added recycle function for pool manager as well - where it recycles the specialized pods
* Switched to rolling update of pods and env variable based change instead of deleting pods in new deploy executor
