Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Documentation for Event Compression #4372

Merged
merged 1 commit into from
Feb 14, 2015
Merged

Documentation for Event Compression #4372

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
79 changes: 79 additions & 0 deletions docs/design/event_compression.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,79 @@
# Kubernetes Event Compression

This document captures the design of event compression.


## Background

Kubernetes components can get into a state where they generate tons of events which are identical except for the timestamp. For example, when pulling a non-existing image, Kubelet will repeatedly generate ```image_not_existing``` and ```container_is_waiting``` events until upstream components correct the image. When this happens, the spam from the repeated events makes the entire event mechanism useless. It also appears to cause memory pressure in etcd (see [#3853](https://github.com/GoogleCloudPlatform/kubernetes/issues/3853)).

## Proposal
Each binary that generates events (for example, ```kubelet```) should keep track of previously generated events so that it can collapse recurring events into a single event instead of creating a new instance for each new event.

Event compression should be best effort (not guaranteed). Meaning, in the worst case, ```n``` identical (minus timestamp) events may still result in ```n``` event entries.

## Design
Instead of a single Timestamp, each event object [contains](https://github.com/GoogleCloudPlatform/kubernetes/blob/master/pkg/api/types.go#L1111) the following fields:
* ```FirstTimestamp util.Time```
* The date/time of the first occurrence of the event.
* ```LastTimestamp util.Time```
* The date/time of the most recent occurrence of the event.
* On first occurrence, this is equal to the FirstTimestamp.
* ```Count int```
* The number of occurrences of this event between FirstTimestamp and LastTimestamp
* On first occurrence, this is 1.

Each binary that generates events will:
* Maintain a new global hash table to keep track of previously generated events (see ```pkg/client/record/events_cache.go```).
* The code that “records/writes” events (see ```StartRecording``` in ```pkg/client/record/event.go```), uses the global hash table to check if any new event has been seen previously.
* The key for the hash table is generated from the event object minus timestamps/count/transient fields (see ```pkg/client/record/events_cache.go```), specifically the following events fields are used to construct a unique key for an event:
* ```event.Source.Component```
* ```event.Source.Host```
* ```event.InvolvedObject.Kind```
* ```event.InvolvedObject.Namespace```
* ```event.InvolvedObject.Name```
* ```event.InvolvedObject.UID```
* ```event.InvolvedObject.APIVersion```
* ```event.Reason```
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we are not careful, people may put ever-changing vaules into the strings of Reason and/or Message.
We can wait to see if this is actually a problem, but experience suggests it will.
Some options then are:

  • if the event creations rate is too high over a time window, begin collapsing events even if they have the differing Reason and Message.
  • use locality sensitive hashing
  • use structured logging, in the style of https://github.com/Sirupsen/logrus to generate event.Message and event.Reason

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, this is not hypothetical. I think it can easily happen today.
If a pod is repeatedly terminating and being restarted by an RC, and you have a lot of nodes, then you are going to see this Message:
Successfully assigned kibana-logging-controller-$FOO to kubernetes-minion-$BAR.c.saad-dev-vms.internal
a bunch of times with different values of $FOO and $BAR.
So, this needs attention sooner rather than later.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have lots more to say about this topic, which might be better said tomorrow in the office, over a tasty beverage.

* ```event.Message```
* If the key for a new event matches the key for a previously generated events (meaning all of the above fields match between the new event and some previously generated event), then the event is considered to be a duplicate:
* Instead of the usual POST/create event API, the new PUT (update) event API is called to update the existing event entry in etcd with the new last seen timestamp and count.
* The event is also updated in the global hash table with an incremented count, updated last seen timestamp, name, and new resource version (all required to issue a future event update).
* If the key for a new event does not match the key for any previously generated event (meaning none of the above fields match between the new event and any previously generated events), then the event is considered to be new/unique:
* The usual POST/create event API is called to create a new event entry in etcd.
* An entry for the event is also added to the global hash table.

## Issues/Risks
* Hash table clean up
* If the component (e.g. kubelet) runs for a long period of time and generates a ton of unique events, the hash table could grow very large in memory.
* *Future consideration:* remove entries from the hash table that are older than some specified time.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This should be handled now, not in the future.
The cache should be fixed size, not a time-based cleanup.
Consider using an lru cache like https://github.com/hashicorp/golang-lru.

* Event history is not preserved across application restarts
* Each component keeps track of event history in memory, a restart causes event history to be cleared.
* That means that compression will not occur across component restarts.
* Similarly, if in the future events are aged out of the hash table, then events will only be compressed until they age out of the hash table, at which point any new instance of the event will cause a new entry to be created in etcd.

## Example
Sample kubectl output
```
FIRSTTIME LASTTIME COUNT NAME KIND SUBOBJECT REASON SOURCE MESSAGE
Thu, 12 Feb 2015 01:13:02 +0000 Thu, 12 Feb 2015 01:13:02 +0000 1 kubernetes-minion-4.c.saad-dev-vms.internal Minion starting {kubelet kubernetes-minion-4.c.saad-dev-vms.internal} Starting kubelet.
Thu, 12 Feb 2015 01:13:09 +0000 Thu, 12 Feb 2015 01:13:09 +0000 1 kubernetes-minion-1.c.saad-dev-vms.internal Minion starting {kubelet kubernetes-minion-1.c.saad-dev-vms.internal} Starting kubelet.
Thu, 12 Feb 2015 01:13:09 +0000 Thu, 12 Feb 2015 01:13:09 +0000 1 kubernetes-minion-3.c.saad-dev-vms.internal Minion starting {kubelet kubernetes-minion-3.c.saad-dev-vms.internal} Starting kubelet.
Thu, 12 Feb 2015 01:13:09 +0000 Thu, 12 Feb 2015 01:13:09 +0000 1 kubernetes-minion-2.c.saad-dev-vms.internal Minion starting {kubelet kubernetes-minion-2.c.saad-dev-vms.internal} Starting kubelet.
Thu, 12 Feb 2015 01:13:05 +0000 Thu, 12 Feb 2015 01:13:12 +0000 4 monitoring-influx-grafana-controller-0133o Pod failedScheduling {scheduler } Error scheduling: no minions available to schedule pods
Thu, 12 Feb 2015 01:13:05 +0000 Thu, 12 Feb 2015 01:13:12 +0000 4 elasticsearch-logging-controller-fplln Pod failedScheduling {scheduler } Error scheduling: no minions available to schedule pods
Thu, 12 Feb 2015 01:13:05 +0000 Thu, 12 Feb 2015 01:13:12 +0000 4 kibana-logging-controller-gziey Pod failedScheduling {scheduler } Error scheduling: no minions available to schedule pods
Thu, 12 Feb 2015 01:13:05 +0000 Thu, 12 Feb 2015 01:13:12 +0000 4 skydns-ls6k1 Pod failedScheduling {scheduler } Error scheduling: no minions available to schedule pods
Thu, 12 Feb 2015 01:13:05 +0000 Thu, 12 Feb 2015 01:13:12 +0000 4 monitoring-heapster-controller-oh43e Pod failedScheduling {scheduler } Error scheduling: no minions available to schedule pods
Thu, 12 Feb 2015 01:13:20 +0000 Thu, 12 Feb 2015 01:13:20 +0000 1 kibana-logging-controller-gziey BoundPod implicitly required container POD pulled {kubelet kubernetes-minion-4.c.saad-dev-vms.internal} Successfully pulled image "kubernetes/pause:latest"
Thu, 12 Feb 2015 01:13:20 +0000 Thu, 12 Feb 2015 01:13:20 +0000 1 kibana-logging-controller-gziey Pod scheduled {scheduler } Successfully assigned kibana-logging-controller-gziey to kubernetes-minion-4.c.saad-dev-vms.internal

```

This demonstrates what would have been 20 separate entries (indicating scheduling failure) collapsed/compressed down to 5 entries.

## Related Pull Requests/Issues
* Issue [#4073](https://github.com/GoogleCloudPlatform/kubernetes/issues/4073): Compress duplicate events
* PR [#4157](https://github.com/GoogleCloudPlatform/kubernetes/issues/4157): Add "Update Event" to Kubernetes API
* PR [#4206](https://github.com/GoogleCloudPlatform/kubernetes/issues/4206): Modify Event struct to allow compressing multiple recurring events in to a single event
* PR [#4306](https://github.com/GoogleCloudPlatform/kubernetes/issues/4073): Compress recurring events in to a single event to optimize etcd storage