Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Makes kafka a feature folder too, consistent with the rest | Staffan Olsson | 2017-11-09 | 1 | -87/+0 |
| | |||||
* | Updates statefulset to beta2 workload API, k8s 1.8 | Staffan Olsson | 2017-11-09 | 1 | -1/+5 |
| | |||||
* | Merge pull request #81 from Yolean/kafka-readiness-tcp | solsson | 2017-11-09 | 1 | -5/+3 |
|\ | | | | | Use Kubernetes native TCP readiness probe for kafka | ||||
| * | Actually, for readiness not liveness, let's go offline fast ...kafka-readiness-tcp | Staffan Olsson | 2017-11-07 | 1 | -1/+1 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | and trust alarms on Under-replicated Partitions to let us know when something is really wrong. Do clients actually care about Readiness? The bootstrap service (https://github.com/Yolean/kubernetes-kafka/pull/52) will definitely care, which is good. The `broker` service, that the StatefulSet manifest depends on (https://github.com/Yolean/kubernetes-kafka/blob/v2.1.0/50kafka.yml#L7) for naming, is without `publishNotReadyAddresses`. Clients will bootstrap, get the individual DNS names of brokers, resolve those addresses and connect directly to pods. | ||||
| * | Timeout was 1s with the old /bin/sh exec, now it's interval -1 s | Staffan Olsson | 2017-11-07 | 1 | -1/+1 |
| | | |||||
| * | More lightweight kafka probe, essentially the same thing | Staffan Olsson | 2017-10-16 | 1 | -5/+3 |
| | | |||||
* | | Uses the new storage class field, requires 1.8 | Staffan Olsson | 2017-11-09 | 1 | -2/+1 |
| | | |||||
* | | Adds storage class for kafka brokers | Staffan Olsson | 2017-11-09 | 1 | -0/+2 |
| | | |||||
* | | Merge pull request #78 from Yolean/outside-services | solsson | 2017-11-07 | 1 | -1/+12 |
|\ \ | | | | | | | Outside (i.e. cluster-external) access through per-broker Service | ||||
| * | | Adds port names as documentation in kafka manifestoutside-services | Staffan Olsson | 2017-10-25 | 1 | -2/+4 |
| | | | |||||
| * | | Exemplifies external access using host lookup that works with minikube | Staffan Olsson | 2017-10-15 | 1 | -0/+1 |
| | | | | | | | | | | | | | | | | | | Listener ports from https://hub.docker.com/r/wurstmeister/kafka/ Config based on KIP-103 | ||||
| * | | Gives init container access to pod name and namespace | Staffan Olsson | 2017-10-15 | 1 | -0/+8 |
| |/ | |||||
* / | Runs Kafka 1.0.0v2.1.0 | Staffan Olsson | 2017-11-01 | 1 | -1/+1 |
|/ | |||||
* | Shares debian with kafka-initutils -> faster first start | Staffan Olsson | 2017-10-15 | 1 | -1/+1 |
| | |||||
* | initutils based on latest debian slim | Staffan Olsson | 2017-10-15 | 1 | -1/+1 |
| | |||||
* | ./update-kafka-image.sh to 0.11.0.1 | Staffan Olsson | 2017-10-02 | 1 | -1/+1 |
| | |||||
* | We prefer Ready:False status instead of restarted pods, | Staffan Olsson | 2017-08-05 | 1 | -1/+1 |
| | | | | | | at least for now, as it allows exec into the pods to investigate. We've been having frequent restarts that are not due to OOMKilled (i.e. not #49). Now failed probes will lead to unready pods, which we can monitor for using #60. | ||||
* | I suppose most init scripts will do fine with curl+kubectl+bash | Staffan Olsson | 2017-08-05 | 1 | -1/+1 |
| | |||||
* | Uses an image with kubectl based on the same debian tag as kafka | Staffan Olsson | 2017-08-05 | 1 | -1/+1 |
| | |||||
* | Looks up zone if kubectl is found, tries to not break config otherwise | Staffan Olsson | 2017-08-05 | 1 | -0/+5 |
| | |||||
* | Makes /metrics export opt-in (through addon branch coming up) | Staffan Olsson | 2017-07-28 | 1 | -15/+0 |
| | |||||
* | With no bash tricks in command we can use the actual bin ... | Staffan Olsson | 2017-07-26 | 1 | -9/+10 |
| | | | | avoiding a layer of indirection and the associated gotchas with signals. | ||||
* | Moves broker.id config into init script | Staffan Olsson | 2017-07-26 | 1 | -2/+1 |
| | |||||
* | Demonstrates how an init script can be used to ... | Staffan Olsson | 2017-07-26 | 1 | -0/+7 |
| | | | | | modify config prior to kafka start. My aim with this is to get rid of the /bin/sh startup scripts and use bin + args directly. | ||||
* | Default shell on Debian shows the same symptom ... | Staffan Olsson | 2017-07-26 | 1 | -1/+1 |
| | | | | | | | | of not forwarding signals as Alpine did. Kafka logs say nothing, and after 30s the container is terminated. With /bin/bash instead the log indicates shutdown behavior. This reverts commit c188f43cb8a252cd685a4944d35577ebc17a3668. | ||||
* | With explicit log4j path we can change config mount ... | Staffan Olsson | 2017-07-26 | 1 | -3/+3 |
| | | | | | | to not mix with sample config. See https://github.com/solsson/dockerfiles/pull/10. | ||||
* | Tagged with the policy from https://github.com/solsson/dockerfiles/pull/11 | Staffan Olsson | 2017-07-26 | 1 | -1/+1 |
| | |||||
* | Clarifies a gotcha: to mount config with log4j.properties ... | Staffan Olsson | 2017-07-26 | 1 | -1/+3 |
| | | | | | | | you must use /opt/kafka/config, due to how log4j.properites (sometimes tools- or connect-) are resolved by the ./bin scripts. See https://github.com/solsson/dockerfiles/pull/10 | ||||
* | New build at commit 0314080 | Staffan Olsson | 2017-07-26 | 1 | -1/+1 |
| | |||||
* | New build with https://github.com/solsson/dockerfiles/pull/9 | Staffan Olsson | 2017-07-25 | 1 | -1/+1 |
| | |||||
* | I think time saved by auto-creating topics will be lost ... | Staffan Olsson | 2017-07-25 | 1 | -0/+1 |
| | | | | tenfold in confusion caused by typos etc | ||||
* | With stock config we have to change zookeeper lookup from the default localhostkafka-011-config-map | Staffan Olsson | 2017-07-25 | 1 | -0/+1 |
| | |||||
* | Use config map's config instead of image's | Staffan Olsson | 2017-07-25 | 1 | -0/+6 |
| | |||||
* | Default shell on debian should forward signals properly | Staffan Olsson | 2017-07-23 | 1 | -1/+1 |
| | |||||
* | Upgrades to current https://github.com/solsson/dockerfiles/pull/5 | Staffan Olsson | 2017-07-23 | 1 | -1/+1 |
| | |||||
* | Upgrades to latest build from https://github.com/solsson/dockerfiles/pull/4, ↵ | Staffan Olsson | 2017-06-28 | 1 | -1/+1 |
| | | | | with plain logging>=INFO config | ||||
* | Raises memory limit for metrics; got 10 OOMKilled per pod in the last 3 hours | Staffan Olsson | 2017-06-27 | 1 | -1/+1 |
| | |||||
* | Adds probes, but for Kafka I don't think it indicates readiness... | Staffan Olsson | 2017-06-27 | 1 | -0/+6 |
| | | | | | | | | | | | | | | | which might not matter because we no longer have a loadbalancing service. These probes won't catch all failure modes, but if they fail we're pretty sure the container is malfunctioning. I found some sources recommending ./bin/kafka-topics.sh for probes but to me it looks risky to introduce a dependency to some other service for such things. One such source is https://github.com/kubernetes/charts/pull/144 The zookeeper probe is from https://kubernetes.io/docs/tasks/configure-pod-container/configure-liveness-readiness-probes/ An issue is that zookeeper's logs are quite verbose for every probe. | ||||
* | Reverts to default termination period, and uses bash for "shell form"... | Staffan Olsson | 2017-06-27 | 1 | -2/+2 |
| | | | | | | | | as Alpine's /bin/busybox (ash) does not forward signals, according to https://pracucci.com/graceful-shutdown-of-kubernetes-pods.html The reason for the termination period change is that we haven't observed any termination behavior yet so we can't know how slow it might be. | ||||
* | Got quite repeatable OOMKilled on pzoo pods, so I figured it must be...resource-limits | Staffan Olsson | 2017-06-27 | 1 | -1/+1 |
| | | | | in metrics becuase nither zoo nor kafka has limits | ||||
* | A monitoring-only pod uses 0m / ~32Mi resources | Staffan Olsson | 2017-06-27 | 1 | -1/+8 |
| | |||||
* | Adds tentative resource requests, based on what idle pods use (though this ↵ | Staffan Olsson | 2017-06-27 | 1 | -0/+4 |
| | | | | includes monitoring) | ||||
* | Uses dynamically provisioned volume for Kafka too. It has matured, ... | Staffan Olsson | 2017-06-27 | 1 | -2/+2 |
| | | | | | simplifies use of multi-zone clusters, works in Minikube, and volumes are deleted when the PV is deleted (by manual kubectl). | ||||
* | Uses the same data path convention as zookeeper, from Confluent Platform | Staffan Olsson | 2017-06-25 | 1 | -2/+2 |
| | |||||
* | Use a specific build, ↵metrics-via-jmx | Staffan Olsson | 2017-06-25 | 1 | -1/+1 |
| | | | | github.com/solsson/dockerfiles/commit/81e8e4c20be19835d4036db1c975ff17455fe7e7 | ||||
* | Adds metrics exporter for Prometheus | Staffan Olsson | 2017-06-25 | 1 | -0/+10 |
| | |||||
* | Support for generic image results in a verbose startup command, make it git ↵ | Staffan Olsson | 2017-06-25 | 1 | -1/+6 |
| | | | | friendly | ||||
* | Default log dir is /tmp/kafka-logs so it should be changed to be inside the ↵ | Staffan Olsson | 2017-06-25 | 1 | -1/+1 |
| | | | | persistent volume mount | ||||
* | Uses the new small image, with only selected jars from the build | Staffan Olsson | 2017-06-25 | 1 | -1/+1 |
| | |||||
* | Borrows string trick from https://github.com/kubernetes/charts/ | Staffan Olsson | 2017-06-25 | 1 | -1/+1 |
| |