The transport in RoundTripper is the default transport and there's just one object shared between different http requests and releasing the idle Connections ended up releasing the idle connections that were established with the executor service address too.
So every time a http request was made to get a service for function, the transport layer ended up having to dial a new connection (that was timing out due to various reasons).
By removing the CloseIdleConnections, we minimize the need for transport to dial a new connection to executor for every request, thereby minimizing the occurrence of dial timeouts.
The router's cache entry for a function might become stale if the pod that had the function specialized gets deleted somehow. In such a case, we'd retry getting a new service for the function from executor and retry forwarding the user request to the newly created service.
This the very first step for fission to integrate with Istio, which is an open platform to connect, manage, and secure microservices. With Istio, users are able to monitor functions usage and trace requests latency through dashboards. For more information, please visit http://fission.io/docs/
A newdeploy backend which uses new deployment to serve requests. This is the second phase of #193 and builds on top of changes in #384 .
* Executor layer added on top of pool manager
* Removed the external server for executor
* Minor changes to keep existing semantics as much possible
* Separating the executor vs. poolmgr backend functionality and associated data members
* Executor logic separated from Poolmgr backend completely, placeholder for new backend
* Changed references to poolmgr in tests
* Moved poolmgr to it's package, as a side effect moved Cache to its's package (was causing cyclical dependency) and had to make some data structures exposed outside package
* Rebased from master and changed references to tpr -> crd
* Executor layer added on top of pool manager
* Executor logic separated from Poolmgr backend completely, placeholder for new backend
* Changed podName to a generic objectReference in fscache (#391)
Changed podName to a generic objectReference in function service cache implementation.
* Moved poolmgr to it's package, as a side effect moved Cache to its's package (was causing cyclical dependency) and had to make some data structures exposed outside package
* Rebased from master and changed references to tpr -> crd
* Merged from master with latest changes
* Executor layer added on top of pool manager
* Removed the external server for executor
* Minor changes to keep existing semantics as much possible
* Separating the executor vs. poolmgr backend functionality and associated data members
* Executor logic separated from Poolmgr backend completely, placeholder for new backend
* Changed references to poolmgr in tests
* update compiling.md to use helm
* Compile instructions: changed pullPolicy to IfNotPresent (#378)
Containers will get stuck in ErrImagePull/ImagePullBackOff state otherwise
* Moved poolmgr to it's package, as a side effect moved Cache to its's package (was causing cyclical dependency) and had to make some data structures exposed outside package
* Fetcher called when pod is created for newDeploy backend but also supports older way, this is WIP and still needs pod specialization and creating & exposing a service so the URL can be hit by end user
* WIP Specializing the POD as part of startup along with fetching
* Working specialization of a new deployment. Needs some work on caching, cleanup etc.
* Switched to service based address instead of POD address
* Minor formating issue fixed
* Added logging to pods and a readiness check, the readiness check is flaky though ATM
* Fixed some rebase issues that were failing build
* Better names for K8S objects and methods
* Switched usage of FuncSvc in backends from pod to api.ObjectReference
* Adding retry to fetcher request, for now just using default retry client which might need tweaking in future
* Switching to plain old retry, some issue in getting retryablehttp with glide import
* Removed stale executor service & deployment from previous merge
* Addressed review comments, still testing some areas
* Added types in FunctionSpec
* Resolved conflicts due to merge from executor_abstraction branch
* Added backend type on EnvironmentSpec along with operations for create/list/update, the pools are created/destroyed based on change in backend type
* Backend from types and a minor err return issue fixed
* Draft version of CPU and memory parameters added to environment
* Added resourceReq to newDeploy, though it has some issues
* Issue with resourceName fixed, now newdeploy pods also pick up resources from the environment config
* Adding scale params, removing validation on CPU params for now
* Fixed a formatting issue
* Checking if slight more delay helps in the test which is currently failing for internal routes
* The resourceList newly added in Env can not be compared by compiler, hence must use breakdown comparison instead
* Added strategy selection on client side
* Added caching, informers, delete operations for newdeploy backend functions
* Deleted a stale directory
* A simple HPA based on scale parameters, testing still WIP
* Fixed a small issue in delete function, added HPA delete too when deleting a function
* Previous merge missed the pkg flag for update fn command somehow, fixed that
* Fixed comments from review
* Changed poolmgr cleanup to be generic cleanup and moved to executor, added instanceID labels to newdeploy so that cleanup works
* Moved instanceIdLabel to types to avoid cyclic dependency
* More review fixes
* Tweaking sleep to see results
* If user does not provide poolsize, then it should not default to zero
* Switched to naming convention for now, fixed default poolsize if not provided
* Changed error return behaviour in delete fn, also changed cleanup to look based on obj type though support for additional type will need more work
* Changed check location so avoid false logging
* Test for newdeploy backend
* Adding tests for poolmgr backend
* Fixed an issue with glide dependency version, already fixed in master
* Added instanceId for NewDeploy, Initial cleanup now cleans older objects of newdeploy backend, removed eagercreate flag and instead using minScale to drive eager creation
* Moved cleanup to executor layer with cleanup for newDeploy backend, changes to use the new Cache impl
* Cleaning up pod & rs along with deployment for newdeploy backend
* Enhanced fn and env listing to show min/maxscale and resuorces respectively
* Added conditional heapster deployment and fixed a small issue with resources for fetcher container in function pod
* Addressed review comments from previous change
* Addressed some more review comments - majorly create only on NotFoundError
* Added TargetCPU as an input for scaling
* Bumped target CPU to be greater than 0 and added a default value
* Min replicas should be 1 even if the minScale is 0 when creating deployment
* Changed name from 'backend' to executorType, added additional test for minscale 0 case, changed TargetCPU to TargetCPUPercent
This change adds a layer of abstraction over poolmgr. Poolmgr is now just one of the ways to turn a function into a service; other implementations will be added. The executor abstraction is a uniform API over all these implementations.
* Executor layer added on top of pool manager
* Removed the external server for executor
* Minor changes to keep existing semantics as much possible
* Separating the executor vs. poolmgr backend functionality and associated data members
* Executor logic separated from Poolmgr backend completely, placeholder for new backend
* Changed references to poolmgr in tests
* Moved poolmgr to it's package, as a side effect moved Cache to its's package (was causing cyclical dependency) and had to make some data structures exposed outside package
* Rebased from master and changed references to tpr -> crd
* Executor layer added on top of pool manager
* Executor logic separated from Poolmgr backend completely, placeholder for new backend
* Changed podName to a generic objectReference in fscache (#391)
Changed podName to a generic objectReference in function service cache implementation.
* Moved poolmgr to it's package, as a side effect moved Cache to its's package (was causing cyclical dependency) and had to make some data structures exposed outside package
* Rebased from master and changed references to tpr -> crd
* Merged from master with latest changes
* Removed stale executor service & deployment from previous merge
* Addressed review comments, still testing some areas
This fixes a bug where functionReferenceResolver returned out-of-date function metadata and caused the router to proxy requests to old function pods.
It also uses the go context package to shutdown the controller when the router is shutting down.
Switch Fission's storage over to the new CustomResourceDefinitions, from the deprecated ThirdPartyResources. This allows us to be compatible with Kubernets 1.8 and onwards.
This also adds a CLI tool for dumping state from an old fission version and restoring state into new CRDs.
The storage service is unaffected by this change.
Add a flag to the environment to control multiple specialization -- the max number of functions per container. This can be set to 1 or infinity.
Add an api proxy to workflow apiserver from the controller.
Add function metadata to FunctionLoadRequest; every v2 environment now knows which function it's loading (but can ignore that information if it wants to).
Add function identity headers to router. This is useful for multiple specialization, so the router can disambiguate between different function calls. (If this turns out to be a non-trivial perf overhead, we could add these headers conditionally, but for now they are always added.)
Watch functions in the router and use that to trigger functionReferenceResolver cache invalidation.
We might want to rate-limit syncTriggers in a future change, since we might be triggering it more often than needed.
This changes the core fission function, environment and trigger types. It also changes Fission's storage to use ThirdPartyResources.
- Functions are now specified by packages. Functions can also have both source and deployment packages. A package can be specified by a literal, or by a URL.
- Environments have a build and runtime component.
- Triggers reference functions by a FunctionReference. This is a layer of indirection between triggers and functions, and will allow things like incremental function upgrades in future releases.
See Documentation/wip/env-v2.md for design discussion about points 1 and 2.
Changes:
* V2 Types
All types now have a spec, following the pattern of K8s objects.
Functions now have source and deployment packages. A Package can be
specified by literal, or by URL.
Environments now have a builder and runtime component.
All triggers use a new FunctionReference to specify the function. This
for now only uses a function name, but in the future can be extended
to be more flexible.
A new FunctionLoadRequest type is added for specialization requests to
the environment runtime.
* TPR types, TPR init code, and a "fission client"
Implements TPR types using the spec types in fission/types.go.
Adds code for adding creating TPR types, and convenient types for crud
operations on each of our resource types.
Adds code for connecting to K8s API and configuring a REST client with
fission types set up.
* Change old stateful controller into a thin apiserver
This apiserver is now simply a stateless api layer on top of the TPR
types. At the moment it doesn't do anything that couldn't be done by
simply talking to the TPR types. In the future we can have better
validation and potentially some higher level APIs (like versioning for
example) in here.
* Split controller client into files and update for v2 types.
* Update CLI for v2 types.
As far as possible we keep the CLI flags the same. We'll have to add
flags for source/deploy packages and builder/runtime
environments. That will come in the next change.
* Update poolmgr and fetcher for v2 types.
Also adds a poolmgr_test.
* Update router for new types.
Also adds a function reference resolver, which separates out the job
of resolving a FunctionReference to a function.
* Update kubewatcher and timer for v2 types.
* Update Message Queue trigger type for v2 types.
* Minor odds and ends.
* Fission bundle CLI updates
Remove controllerUrl flag, since we don't need it any more.
* Remove etcd deployment (replaced by storing state in TPR)
Also update the poolmgr commandline, and use an env var for the
fetcher image URL.
* Explicit ChecksumType and consts
* Clarify separation of environment interface types
Addresses #158. Adds URL params into HTTP headers. A URL param named `KEY` with the value `VALUE` in the actual URL will appear as the header "X-Fission-Params-KEY: VALUE" in the fission function.
The poolmgr client now aggregates successive tapService() calls and sends them to the poolmgr service every 5 sec, instead of sending them immediately.
Before this change, if the user doesn't define a route for /, the
router happily 404s at that path. This breaks GKE Ingress, which uses
that request as a health check. So this change adds a handler for
"GET /", unless the user already has one.
Also, log the delay for slow requests.
Poolmgr removed the default-latest-version semantics to simplify
caching. But this broke the router's internal function routes, which
are used by kubeWatcher. The router now exposes a versionless url
and regularly updates that to point to the latest version.
Addresses issue #51. Makes versioning in poolmgr explicit, so its
cache needs no invalidation on version update. Functions in poolmgr
are always cached by name and UID.
Router now updates implictly versioned routes with the latest version
of a function.
This means that users will see requests to implicitly-versioned routes
go to the latest version of a function within 3 seconds. (Those 3 sec
will go away when we use a real watch instead of polling the
controller.)
There's no change in behaviour for routes that explictly specify a
function UID.
These routes allow us to call any function without explicitly defining
a route for them. TODO: if we can serve these routes from a separate
instance of router (controlled by a commandline flag or env var), we'd
be able to keep this instance of the router "private" by not assigning
it a loadbalancer/nodeport or other externally-visible service.
Poolmgr needs to know usage statistics for a function's pod. This
change asynchronously taps poolmgr API when router uses a service.
Poolmgr can use this information to control pod expiry. It may also
be useful later as one of the metrics for autoscaling.
Create an implementation of http.RoundTrip which does retries --
RetryingRoundTripper. K8s services seem to timeout for about ~1-2 sec
after they're created even when the pods being routed to are ready to
serve requests.
Router uses poolmgr to specialize pods when necessary.
Router uses controller to get the list of triggers to listen for. For
now this integration is pretty crappy -- we just poll the controller
every few seconds and cache the result. The right way would be to
have some sort of watch API on the controller and use that. Or maybe
share access to etcd directly.
