Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/tleyden/deepstyle

Run DeepStyle (aka NeuralStyle) in the cloud
https://github.com/tleyden/deepstyle

Last synced: 3 months ago
JSON representation

Run DeepStyle (aka NeuralStyle) in the cloud

Awesome Lists containing this project

README

        

## Overview

This is the backend "changes worker" for the [deepstyle-ios](https://github.com/tleyden/deepstyle-ios/) app.

[Sync Gateway](https://waffle.io/couchbase/sync_gateway) provides the REST api, and [Couchbase Server](https://github.com/couchbase/manifest) provides the persistence layer.

* Sync Gateway changes listener to track the number of unprocessed jobs
* When new jobs are detected, publishes stats to CloudWatch
* CloudWatch Alarms cause AWS EC2 instances with GPU processors to spin up
* Look for unprocessed jobs
* Invoke [neural-style](https://github.com/jcjohnson/neural-style) to apply the artistic style to the photograph
* Update job with results, which will sync down to the [deepstyle-ios](https://github.com/tleyden/deepstyle-ios/) app

## Steps to run

* Kick off Sync Gateway running [this config](https://github.com/tleyden/deepstyle/blob/master/docs/sync-gateway-config.json)
* Kick off ami `ami-5587c93f` (private AMI at the moment, stay tuned)
* Run `deepstyle follow_sync_gw --url http://demo.couchbasemobile.com:4984/deepstyle/`
* Use Paw/Curl to upload images

## JSON Docs

### Job

```
{
"_attachments":{
"photo":{
"content_type":"image/png",
"digest":"sha1-U8DAjp4S6T4HWWfo+HAdULGZpmw=",
"length":3479844,
"revpos":11,
"stub":true
},
"style_image":{
"content_type":"image/png",
"digest":"sha1-IvVjV5TaT57zICYRpl/NJqJYVds=",
"length":283092,
"revpos":13,
"stub":true
}
},
"_id":"job",
"_rev":"13-17ec2f655e6fc8e41809a00df7b15e53",
"state":"READY_TO_PROCESS",
"type":"job"
}
```

### Job States

* NOT_READY_TO_PROCESS (no attachments yet)
* READY_TO_PROCESS (attachments added)
* BEING_PROCESSED (worker running)
* PROCESSING_SUCCESSFUL (worker done, added result attachment)
* PROCESSING_FAILED (worker done, added error msg)

## Job Queue Processor

* For each change where type=job and state=READY_TO_PROCESS:
* Change state to BEING_PROCESSED and update doc
* Download attachments to temp files
* Kick off exec and tell it to store result in a temp file
* Wait for exec to finish
* Add new attachment to doc with result
* Change state to PROCESSING_SUCCESSFUL (or failed if exec failed)
* Delete temp files

## Adding a new command (cobra)

```
cobra add publish_cloudwatch_metrics
```

## Creating cloudwatch alarms to trigger autoscale

```
$ aws autoscaling put-scaling-policy --policy-name deepstyle-scalout --auto-scaling-group-name DeepStyle --scaling-adjustment 1 --adjustment-type ChangeInCapacity --profile tleyden
$ aws autoscaling put-scaling-policy --policy-name deepstyle-scalein --auto-scaling-group-name DeepStyle --scaling-adjustment -1 --adjustment-type ChangeInCapacity --profile tleyden
$ aws cloudwatch put-metric-alarm --alarm-name AddCapacityToProcessDeepStyleQueue3 --metric-name NumJobsReadyOrBeingProcessed --namespace "DeepStyleQueue" --statistic Average --period 60 --threshold 1 --comparison-operator GreaterThanOrEqualToThreshold --evaluation-periods 1 --alarm-actions $SCALE_OUT_ARN --profile tleyden
$ aws cloudwatch put-metric-alarm --alarm-name RemoveCapacityToProcessDeepStyleQueue3 --metric-name NumJobsReadyOrBeingProcessed --namespace "DeepStyleQueue" --statistic Average --period 60 --threshold 0 --comparison-operator LessThanOrEqualToThreshold --evaluation-periods 1 --alarm-actions $SCALE_IN_ARN --profile tleyden

```