* Define a new telemetry Stat API
Proposal definition for a new Stat API, for the purposes of satisfying the queries proposed in #627.
StatSummary will replace Stat once implemented and the original Stat deleted.
* Extracted logic from destination server
* Make tests follow style used elsewhere in the code
* Extract single interface for resolvers
* Add tests for k8s and ipv4 resolvers
* Fix small usability issues
* Update dep
* Act on feedback
* Add pod-based metric_labels to destinations response
* Add documentation on running control plane to BUILD.md
Signed-off-by: Phil Calcado <phil@buoyant.io>
* Fix mock controller in proxy tests (#656)
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
* Address review feedback
* Rename files in the destination package
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
simulate-proxy uses a deployment object from kubernetes to simulate
each proxy metrics endpoint.
Modify simulate-proxy to instead use a pod to simulate each proxy
metrics endpoint. This ensures that each metrics endpoint consistently
represents a pod in kubernetes, including it's namespace, deployment,
and label information.
This change also adds support for:
- a new `metric-ports` flag, default is `10000-10009`.
- `classification`, `pod_name`, and `pod_template_hash` labels
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Extracted logic from destination server
* Make tests follow style used elsewhere in the code
* Extract single interface for resolvers
* Add tests for k8s and ipv4 resolvers
* Fix small usability issues
* Update dep
* Act on feedback
Signed-off-by: Phil Calcado <phil@buoyant.io>
simulate-proxy increments a single set of metrics on each iteration, and
also randomizes http status codes, leaving counters unchanged across
several collections.
Modify simuilate-proxy to increment all metrics on each iteration,
provide a 90% success rate, ensure a pod does not call itself, and
increase proxy count from 3 to 10.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Add tests/utils/scripts for running integration tests
Add a suite of integration tests in the `test/` directory, as well as
utilities for testing in the `testutil/` directory.
You can use the `bin/test-run` script to run the full suite of tests,
and the `bin/test-cleanup` script to cleanup after the tests.
The test/README.md file has more information about running tests.
@pcalcado, @franziskagoltz, and @rmars also contributed to this change.
* Create TEST.md file at the root of the repo
* Update based on review feedback
* Relax external service IP timeout for GKE
* Update TEST.md with more info about different types of test runs
* More updates to TEST.md based on review feedback
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
The Prometheus scrape config collects from Conduit proxies, and maps
Kubernetes labels to Prometheus labels, appending "k8s_".
This change keeps the resultant Prometheus labels consistent with their
source Kubernetes labels. For example: "deployment" and
"pod_template_hash".
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Have the controller tell the client whether the service exists, not
just what are available. This way we can implement fallback logic to
alternate service discovery mechanisms for ambigious names.
Signed-off-by: Brian Smith <brian@briansmith.org>
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
The Prometheus config in the docker-compose environment had fallen
behind the prod setup.
This change updates the docker-compose environment in the following
ways:
- Prometheus config more closely matches prod, based on #583
- simulate-proxy labels matches prod, based on #605
- add Grafana container
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The simulate-proxy script pushes metrics to the telemetry service. This PR modifies the script to expose metrics to a prometheus endpoint. This functionality creates a server that randomly generates response_total, request_totals, response_duration_ms and response_latency_ms. The server reads pod information from a k8s cluster and picks a random namespace to use for all exposed metrics.
Tested out these changes with a locally running prometheus server. I also ran the docker-compose.yml to make sure metrics were being recorded by the prometheus docker container.
fixes#498
Signed-off-by: Dennis Adjei-Baah <dennis@buoyant.io>
* Most controller listeners should only bind on localhost
* Use default listening addresses in controller components
* Review feedback
* Revert test_helper change
* Revert use of absolute domains
Signed-off-by: Alex Leong <alex@buoyant.io>
Shortly after conduit is installed in k8s environment. The control plane component that establishes a watch endpoint with k8s run in to networking issues during proxy initialization. During failure, each watcher fails to retry its connection to k8s watch endpoint which leads to timeouts and eventually, multiple controller pod restarts.
This PR adds retry logic to each "watch" enabled package.
fixes#478
Signed-off-by: Dennis Adjei-Baah <dennis@buoyant.io>
When the conduit proxy is injected into the controller pod, we observe controller pod proxy stats show up as an "outbound" deployment for an unrelated upstream deployment. This may cause confusion when monitoring deployments in the service mesh.
This PR filters out this "misleading" stat in the public api whenever the dashboard requests metric information for a specific deployment.
* exclude telemetry generated by the control plane when requesting deployment metrics
fixes#370
Signed-off-by: Dennis Adjei-Baah <dennis@buoyant.io>
The dns_test had assumed DNS changes were deterministically ordered, but
util.DiffAddresses uses a map and therefore does not guarantee ordering.
Fix dns_test to sort TCP Addresses prior to comparison.
Fixes#515
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
When attempting to tap N pods when N is greater than the target rps, a rounding error occurs that requests 0 rps from each pod and no tap data is returned.
Ensure that tap requests at least 1 rps from each target pod.
Tested in Kubernetes on docker-for-desktop with a 15 replica deployment and a maxRps of 10.
Signed-off-by: Alex Leong <alex@buoyant.io>
The control plane is proxied through the Conduit proxy. The Conduit
proxy is based on the base image, and the control plane containers
and the proxy share a networking namespace. This means we don't
need the extra base utilities in the controller images since we can
use the utilties in the proxy image.
This is a step towards building the initial no-networking Conduit CA
pod. Since the Conduit CA will not do any networking of its own, we
networking debugging utilties are not helpful for it. They are
actually an unnecessary risk because they could facilitate the
exfiltration of the private key of the CA. (The Conduit CA pod won't
have the Conduit Proxy injected into it either.)
This also simplifies & slightly speeds up the building of the
controller images. This is a stepping stone towards being able to
build the controller images without `docker build` to improve build
times.
Signed-off-by: Brian Smith <brian@briansmith.org>
* Use Go 1.10.0 to build Go components.
Take advantage of the new build cache in Go 1.10. Future work on improving
build performance will utilize the build cache further.
Signed-off-by: Brian Smith <brian@briansmith.org>
Previously Dockerfile-go-deps was converted from a multi-stage Dockefile
to a single-stage Dockerfile in anticipation of enabling efficient use
of `--cache-from` in CI. However, that resulted in the image ballooning
in size because it contained the Git repo for every package downloaded
by `dep ensure`.
Bring the image back down to the proper size by removing the temporary
files created.
Signed-off-by: Brian Smith <brian@briansmith.org>
The instance cache that powers the ListPods API is stored in memory in the telemetry service. This means that when there are multiple replicas of the telemetry service, each replica will have a distinct, incomplete view of the added pods based on which pods report to that telemetry replica. This causes the data plane bubbles on the dashboard to not all be filled in, and to flicker with each data refresh.
We create a Prometheus counter called reports_total which has pod as a label. Whenever a telemetry service instance receives a report from a pod, it increments reports_total for that pod. This allows us to remove the in-memory instance cache and instead query Prometheus to see if each pod has had a report in the last 30 seconds.
Fixes#337
Signed-off-by: Alex Leong <alex@buoyant.io>
In PR #298 we moved time window parsing (10s => (time.now - 10s,
time.now) down the stack to immediately before the query. This had the
unintended effect of creating parallel latency quantile requests with
slightly different timestamps.
This change parses the time window prior to latency quantile fan out,
ensuring all requests have the same timestamp.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
PR #298 moved summary (non-timeseries) requests to Prometheus' Query
endpoint, with no timestamp provided. This Query endpoint returns a
single data point with whatever timestamp was provided in the request.
In the absense of a timestamp, it uses current server time. This causes
the Public API to return discreet data points with slightly different
timestamps, which is unexpected behavior.
Modify the Public API -> Telemetry -> Prometheus request path to always
require a timestamp for single data point requests.
Fixes#340
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
On my system (i9-7960x running Docker natively in Linux) this regularly saves
over 11 seconds of build time when a file under pkg/ changes and over 1.5
seconds of build time when a file under controller/ changes. Since most
contributors are running Docker in a VM on less powerful computers, the
savings for most contributors should be significantly greater.
I imagine the savings for web/ and cli/ and proxy-init/ are similar, but I
did not measure them.
Signed-off-by: Brian Smith <brian@briansmith.org>
Precompiling pkg/ in an earlier layer saves ~10 seconds of wall clock
time on an incremental build on my machine (i9-7960x) when I update a
file in controller/ such as controller/destination/server.go. This
makes a significant difference in the edit-build-test loop.
Signed-off-by: Brian Smith <brian@briansmith.org>
Previously Dockerfile-go-deps would run `dep ensure` whenever anything in the
source tree changed. Also, because it was a multi-stage Dockerfile it did not
work well with Docker's `--cache-from` feature.
Change Dockerfile-go-deps to only re-run `dep ensure` when Gopkg.{toml,lock}
and/or bin/dep change. Simplify it to a single stage so that it works better
with Docker's `--cache-from` feature.
Signed-off-by: Brian Smith <brian@briansmith.org>
bin/dep verifies the digest of the `dep` downloaded `dep` executable,
whereas previously Dockerfile-go-deps wasn't.
Signed-off-by: Brian Smith <brian@briansmith.org>
The Public APIs stat endpoint copies a slice of values to a slice of
pointers prior to gRPC response. Go's range clause re-uses the same
pointer for each iteration of the loop, causing a slice of {1,2,3}
becoming {3,3,3}.
Fix the range loop to directly reference pointers in the slice of
values, ignoring the range variable. Also add tests to catch this case.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Removed the `method` label from Prometheus, and removed HTTP methods from reports. Removed `StreamSummary` from reports and replaced it with a `u32` count of streams.
Closes#266
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
All requests from the public API service to the Telemetry service were
done serially. In some cases a single request to the public API's Stat
endpoint resulted in 5 serial requests to the Telemetry service.
Make all requests from the Public API to Telemetry concurrent.
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Part of #299
Follow-up from #315.
Now that the UIs don't report per-path metrics, we can remove the path label from Prometheus, the path aggregation and filtering options from the telemetry API, and the path field from the proxy report API.
I've modified the tests to no longer expect the removed fields, and manually verified that Conduit still works after making these changes.
Closes#265
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
Prometheus queries from the Telemetry service were taking seconds or 10s
of seconds.
Optimize these queries:
- Move all summary queries requiring a single point data off of Prometheus'
QueryRange() endpoint, onto Query()
- Set `defaultVectorRange` to 30s, and also use it regardless of time
window
Also add tests for grpc_server and telemetry server
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Fixes#260
This PR updates the web UI to remove the pod detail page, and to remove the links to that page from pod names in metrics tables. It also removes the `pods` option from `conduit stat`, and the `sourcePod` and `targetPod` fields from the controller API proto's `MetricMetadata` message.
I've updated the `conduit stat` tests to reflect these changes, and manually verified the web UI changes.
Closes#261
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
The proxy currently stores latency values in an `OrderMap` and reports every observed latency value to the controller's telemetry API since the last report. The telemetry API then sends each individual value to Prometheus. This doesn't scale well when there are a large number of proxies making reports.
I've modified the proxy to use a fixed-size histogram that matches the histogram buckets in Prometheus. Each report now includes an array indicating the histogram bounds, and each response scope contains a set of counts corresponding to each index in the bounds array, indicating the number of times a latency in that bucket was observed. The controller then reports the upper bound of each bucket to Prometheus, and can use the proxy's reported set of bucket bounds so that the observed values will be correct even if the bounds in the control plane are changed independently of those set in the proxy.
I've also modified `simulate-proxy` to generate the new report structure, and added tests in the proxy's telemetry test suite validating the new behaviour.
We now create a new test HTTP server per test case instead of sharing it across them all.
This should solve the data races we have experienced on Travis.
Signed-off-by: Phil Calcado <phil@buoyant.io>
We previously did not have race detection enabled because our tests
would fail. Following #249, this is no longer the case.
Enable race detection in ci and build instructions. This change also
fixes client_test.go attempting to allocate a 2GB buffer due to bad test
input.
Fixes#173
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
The conduit dashboard command asychronously shells out and runs "kubectl
proxy".
This change replaces the shelling out with calls to kubernetes proxy
APIs. It also allows us to enable race detection in our go tests, as the
shell out code tests did not pass race detection.
Fixes#173
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Add `bin/dep` which fetches a fixed version of `dep` to be used.
* Upgrade from dep 0.3.1 to 0.4.1
* Fix inconsistent Gopkg.lock by checking in the result of `bin/dep ensure`
Signed-off-by: Alex Leong <alex@buoyant.io>
The conduit.io/* k8s labels and annotations we're redundant in some
cases, and not flexible enough in others.
This change modifies the labels in the following ways:
`conduit.io/plane: control` => `conduit.io/controller-component: web`
`conduit.io/controller: conduit` => `conduit.io/controller-ns: conduit`
`conduit.io/plane: data` => (remove, redundant with `conduit.io/controller-ns`)
It also centralizes all k8s labels and annotations into
pkg/k8s/labels.go, and adds tests for the install command.
Part of #201
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
* Add -log-level flag for install and inject commands
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Turn off all CLI logging by default, rename inject and install flags
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Re-enable color logging
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
We added basic prometheus instrumentation, but this only encapsulated basic go metrics and
request counts. This adds latency and response size metrics exporting as well, to the
public-api server, theweb server and the telemetry server.
Since the util function in grpc.go was basically used to wrap the server creation in a prometheus handler, I added the other prometheus constants in there and renamed the file to prometheus.go.
- Add request duration and response size instrumentation to web and public api
- Also add latency monitoring to telemetry service requests
- Rename util/grpc.go to util/prometheus.go
* Set conduit version to match conduit docker tags
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Remove --skip-inbound-ports for emojivoto
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Rename git_sha => git_sha_head
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Switch to using the go linker for setting the version
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Log conduit version when go servers start
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Cleanup conduit script
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Add --short flag to head sha command
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
* Set CONDUIT_VERSION in docker-compose env
Signed-off-by: Kevin Lingerfelt <kl@buoyant.io>
Previously, running `$conduit tap` would return a `Unexpected EOF` error when the server wasn't available. This was due to a few problems with the way we were handling errors all the way down the tap server. This change fixes that and cleans some of the protobuf-over-HTTP code.
- first step towards #49
- closes#106
* Make Eos optional in TapEvent
grpc_status not being set in protobuf is the same as being set to zero,
which is also status OK
Modify TapEvent to include an optional EOS struct
Signed-off-by: Andrew Seigner <siggy@buoyant.io>
Part of #198
* Add Eos to proto & proxy tap end-of-stream events
The proxy now outputs `Eos` instead of `grpc_status` in all end-of-stream tap events. The EOS value is set to `grpc_status_code` when the response ended with a `grpc_status` trailer, `http_reset_code` when the response ended with a reset, and no `Eos` when the response ended gracefully without a `grpc_status` trailer.
This PR updates the proxy. The proto and controller changes are in PR #204.
Part of #198. Closes#202
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
Currently, all "success"/"failure" classifications in the telemetry API are made based on the `grpc-status` trailer. If the trailer is not present, then a request is assumed to have failed. As we start proxying non-gRPC traffic, the controller needs to also be aware of HTTP status codes, so that non-gRPC requests are not assumed to always fail.
I've modified the telemetry API server to classify requests based on their HTTP status codes when the `grpc-status` trailer is not present.
I've also modified the `simulate-proxy` script to generate fake HTTP/2 traffic without the `grpc-status` trailer.
Closes#196
Signed-off-by: Eliza Weisman <eliza@buoyant.io>
Risha Mars
Phil Calçado
Brian Smith
Andrew Seigner
Kevin Lingerfelt
Dennis Adjei-Baah
Alena Varkockova
Alex Leong
Eliza Weisman