nomad-crashloop-detector

nomad-crashloop-detector

nomad-crashloop-detector is a tool meant to detect allocation crash-loops, by consuming the allocation stream from nomad-firehose in RabbitMQ.

Running

The project got build artifacts for linux, darwin and windows in the GitHub releases tab.

A docker container is also provided at seatgeek/nomad-crashloop-detector

Requirements

• Go 1.8

Building

To build a binary, run the following

# get this repo
go get github.com/seatgeek/nomad-crashloop-detector

# go to the repo directory
cd $GOPATH/src/github.com/seatgeek/nomad-crashloop-detector

# build the `nomad-crashloop-detector` binary
make build

This will create a nomad-crashloop-detector binary in your $GOPATH/bin directory.

Configuration

Any NOMAD_* env that the native nomad CLI tool supports are supported by this tool.

• $AMQP_CONNECTION is identical to $SINK_AMQP_CONNECTION, but is for the consuming stream from nomad-firehose
• $AMQP_QUEUE is the RabbitMQ queue to consume the nomad-firehose from.
• $RESTART_COUNT how many restarts to allow within $RESTART_INTERVAL time (example: 5)
• $RESTART_INTERVAL within what time frame $RESTART_COUNT allocation restarts must happen to trigger an notification (example: 5m)
• $NOTIFICATION_INTERVAL how often a notification should happen on a crash-looping allocation (example: 5m)

Sinks

The sink type is configured using $SINK_TYPE environment variable. Valid values are: stdout, kinesis and amqp.

The amqp sink is configured using $SINK_AMQP_CONNECTION (amqp://guest:guest@127.0.0.1:5672/), $SINK_AMQP_EXCHANGE and $SINK_AMQP_ROUTING_KEY environment variables.

The kinesis sink is configured using $SINK_KINESIS_STREAM_NAME and $SINK_KINESIS_PARTITION_KEY environment variables.

The stdout sink do not have any configuration, it will simply output the JSON to stdout for debugging.

Example

Assuming the following setup:

• nomad exchange (type=topic)
• nomad.crash-loop-in queue which is bound to nomad exchange with routing key allocations
• nomad.crash-loop-out queue which is bound to nomad exchange with routing key crash-loop

Running nomad-firehose:

SINK_TYPE=amqp \
SINK_AMQP_CONNECTION="amqp://guest:guest@127.0.0.1:5672/" \
SINK_AMQP_EXCHANGE=nomad \
SINK_AMQP_ROUTING_KEY=allocations \
nomad-firehose allocations

Running nomad-crashloop-detector:

RESTART_COUNT=2 \
RESTART_INTERVAL=5m \
NOTIFICATION_INTERVAL=5m \
SINK_TYPE=amqp \
SINK_AMQP_CONNECTION="amqp://guest:guest@127.0.0.1:5672/" \
SINK_AMQP_EXCHANGE=nomad \
SINK_AMQP_ROUTING_KEY=crash-loop \
AMQP_CONNECTION=$SINK_AMQP_CONNECTION \
AMQP_QUEUE=nomad.crash-loop-in \
nomad-crashloop-detector

The setup will make nomad-firehose send all nomad allocation changes to the nomad exchange, that will forward messages to the nomad.crash-loop-in queue. nomad-crashloop-detector will consume the messages in nomad.crash-loop-in, and when a restart threshold is reached, submit a AMQP job to the nomad exchange, which will redirect the message to nomad.crash-loop-in.

Example crash-loop payload

{
    "LastEvent": {
        "Name": "job.task[0]",
        "AllocationID": "fd4deb1f-405b-93a6-3eb4-a84e0670049d",
        "DesiredStatus": "run",
        "DesiredDescription": "",
        "ClientStatus": "running",
        "ClientDescription": "",
        "JobID": "job",
        "GroupName": "group",
        "TaskName": "task",
        "EvalID": "db0064ab-a44d-e450-4f66-2cabbec536bb",
        "TaskState": "pending",
        "TaskFailed": false,
        "TaskStartedAt": "2017-07-12T13:56:30.932498912Z",
        "TaskFinishedAt": "0001-01-01T00:00:00Z",
        "TaskEvent": {
            "Type": "Restarting",
            "Time": 1499867806677609000,
            "FailsTask": false,
            "RestartReason": "Restart within policy",
            "SetupError": "",
            "DriverError": "",
            "DriverMessage": "",
            "ExitCode": 0,
            "Signal": 0,
            "Message": "",
            "KillReason": "",
            "KillTimeout": 0,
            "KillError": "",
            "StartDelay": 17425840945,
            "DownloadError": "",
            "ValidationError": "",
            "DiskLimit": 0,
            "DiskSize": 0,
            "FailedSibling": "",
            "VaultError": "",
            "TaskSignalReason": "",
            "TaskSignal": ""
        }
    },
    "EventLog": [
        "2017-07-12T15:56:15.401013209+02:00",
        "2017-07-12T15:56:46.677608921+02:00"
    ]
}