Skip to content

delijati/tchotcho

Repository files navigation

pipeline status coverage report

TchoTcho

 (tchotcho)OOOOOoo...
         _____      oo
 _______ ||_||__n_n__I_
 |__T__|-|_T_|_________)>
  oo oo   o ()() ()() o\

Train on EC2 when you are ready with experimenting!

This tool contains bundle of methods to setup a EC2 instance for training.

Install

❯ pip install tchotcho

Requirements

  • Setup a private/ public key (login to EC2, helpful for cloning private repo)
  • Install rsync (copy project to EC2)
  • Install ssh (execute commands remote on EC2)
  • Setup your aws credentials profile via (~/.aws/credentials)

Actions

Available commands:

❯ tchotcho --help
Usage: tchotcho [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  info
  key
  shell
  spot
  stack

Info

Update

Retrieves list of EC2 instances from https://ec2instances.info/ and queries aws to find suitable ami imges for training. It caches the result locally for later usage.

Help:

❯ AWS_PROFILE=dev tchotcho info update --help
Usage: tchotcho info update [OPTIONS]

  Update the GPU info json file

Options:
  --ownerid TEXT     Owner id used (we use ubuntu)  [default: 898082745236;
                     required]

  --namefilter TEXT  AMI filter by name  [default: Deep Learning AMI* 18.04*;
                     required]

  --limit INTEGER    Number of AMI image  [default: 5; required]
  --csv / --no-csv
  --region TEXT      List in region  [required]
  --help             Show this message and exit.

Usage:

❯ AWS_PROFILE=dev env/bin/tchotcho info update
╒═══════════════╤═══════╤══════════════╤═══════════════════════╤═══════╤═══════════╤═════════════╤═══════════════╕
│ name          │   gpu │   gpu_memory │ gpu_model             │   cpu │    memory │ supported   │ price         │
╞═══════════════╪═══════╪══════════════╪═══════════════════════╪═══════╪═══════════╪═════════════╪═══════════════╡
│ p2.16xlarge   │    16 │          192 │ NVIDIA Tesla K80      │    64 │   732     │ True        │ 21.216        │
├───────────────┼───────┼──────────────┼───────────────────────┼───────┼───────────┼─────────────┼───────────────┤
│ p3dn.24xlarge │     8 │          256 │ NVIDIA Tesla V100     │    96 │   768     │ True        │ Not available │
├───────────────┼───────┼──────────────┼───────────────────────┼───────┼───────────┼─────────────┼───────────────┤
│ p2.8xlarge    │     8 │           96 │ NVIDIA Tesla K80      │    32 │   488     │ True        │ 10.608        │
...
╒══════════════╤═══════════════════════════════════════════════╤══════════════════════════╤═══════════════════════╕
│ region       │ name                                          │ date                     │ ami                   │
╞══════════════╪═══════════════════════════════════════════════╪══════════════════════════╪═══════════════════════╡
│ eu-central-1 │ Deep Learning AMI (Ubuntu 18.04) Version 29.0 │ 2020-05-20T14:57:39.000Z │ ami-062a3145bcf312c71 │
├──────────────┼───────────────────────────────────────────────┼──────────────────────────┼───────────────────────┤
│ eu-central-1 │ Deep Learning AMI (Ubuntu 18.04) Version 28.1 │ 2020-05-03T19:46:44.000Z │ ami-061aaaac62de85935 │
├──────────────┼───────────────────────────────────────────────┼──────────────────────────┼───────────────────────┤
│ eu-central-1 │ Deep Learning AMI (Ubuntu 18.04) Version 28.0 │ 2020-04-29T09:32:42.000Z │ ami-0f162c7e9b0e7d6f1 │
├──────────────┼───────────────────────────────────────────────┼──────────────────────────┼───────────────────────┤
│ eu-central-1 │ Deep Learning AMI (Ubuntu 18.04) Version 27.0 │ 2020-03-03T03:13:27.000Z │ ami-09633db638556dc39 │
├──────────────┼───────────────────────────────────────────────┼──────────────────────────┼───────────────────────┤
│ eu-central-1 │ Deep Learning AMI (Ubuntu 18.04) Version 26.0 │ 2019-12-02T22:09:33.000Z │ ami-0dcdcc4bc9e75005f │
╘══════════════╧═══════════════════════════════════════════════╧══════════════════════════╧═══════════════════════╛

List

Use the cached result from update command from ~/.tchotcho/gpu_info.json.

Key

Function to manage keys on EC2.

❯ AWS_PROFILE=dev tchotcho key --help
Usage: tchotcho key [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  create
  delete
  fingerprint
  import
  list

Shell

Currently rsync and ssh are implemented.

Rsync

Sync your local experiment to EC2. The folder .git is excluded and all rules are applied from .gitignore.

Usage:

❯ tchotcho shell rsync --src . --dst [email protected]:/home/ubuntu/project --privat-key ~/.ssh/id_rsa

Ssh

Execute remote command on EC2.

Usage:

❯ tchotcho shell ssh --host [email protected] --privat-key ~/.ssh/id_rsa --cmd "pwd && date"

Spot

List spot price for given instances and regions.

❯ AWS_PROFILE=dev tchotcho spot list --help
Usage: tchotcho spot list [OPTIONS]

Options:
  --region TEXT     List spot prices in region
  --gpu / --no-gpu  Limit results to show only GPU instances
  --inst TEXT       List spot prices for instance type
  --csv / --no-csv
  --help            Show this message and exit.

List

List spot price by default only supported GPU instances are shown.

Usage:

❯ AWS_PROFILE=dev tchotcho spot list --region eu-central-1
╒════════════════╤════════════════════╤═════════════╕
│ InstanceType   │ AvailabilityZone   │   SpotPrice │
╞════════════════╪════════════════════╪═════════════╡
│ g4dn.xlarge    │ eu-central-1b      │      0.1974 │
├────────────────┼────────────────────┼─────────────┤
│ g4dn.xlarge    │ eu-central-1a      │      0.1974 │
├────────────────┼────────────────────┼─────────────┤
│ g3.4xlarge     │ eu-central-1b      │      0.4275 │
├────────────────┼────────────────────┼─────────────┤
│ g3.4xlarge     │ eu-central-1a      │      0.4275 │
...
│ g3.16xlarge    │ eu-central-1a      │      1.71   │
├────────────────┼────────────────────┼─────────────┤
│ g4dn.16xlarge  │ eu-central-1b      │      1.8593 │
├────────────────┼────────────────────┼─────────────┤
│ p3.8xlarge     │ eu-central-1b      │      4.5876 │
├────────────────┼────────────────────┼─────────────┤
│ p3.8xlarge     │ eu-central-1a      │      4.5876 │
├────────────────┼────────────────────┼─────────────┤
│ p3.16xlarge    │ eu-central-1b      │      9.1752 │
├────────────────┼────────────────────┼─────────────┤
│ p2.8xlarge     │ eu-central-1b      │     10.608  │
├────────────────┼────────────────────┼─────────────┤
│ p2.8xlarge     │ eu-central-1a      │     10.608  │
╘════════════════╧════════════════════╧═════════════╛

Stack

Core of this tool create a EC2 instance via cloudformation.

❯ AWS_PROFILE=dev tchotcho stack --help
Usage: tchotcho stack [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  create
  delete
  list

Create

Create a cloudformation stack. The name needs to be the same as the imported key in aws.

Usage:

❯ AWS_PROFILE=dev tchotcho stack create --help
Usage: tchotcho stack create [OPTIONS]

Options:
  --name TEXT            Name of stack to create && key  [required]
  --ami TEXT             Name of ami to use  [default: ami-061aaaac62de85935;
                         required]

  --inst TEXT            Name of the instance to use  [required]
  --security_group TEXT  Name of the security group to use
  --subnet TEXT          Name of the subnet to use
  --price FLOAT          Name of the instance to use
  --size INTEGER         Size of the disk in GB
  --dry / --no-dry       Only print yaml no create
  --help                 Show this message and exit.

Example:

❯ AWS_PROFILE=dev tchotcho stack create --name test-dl --inst t2.medium --price 0.02 --dry

User script

TODO Currently only via the api changeable. Explain:

  • waiting for apt via flock
  • signal to cloudformation so we wait until user script install is finished

Code style

We use flake8 to ensure code quality and black to autoformat code.

Add it as a git-hook:

$ flake8 --install-hook git
$ git config --bool flake8.strict true

TODO

  • Try paramiko + scp to be pure python

About

Train on ec2 when you are done experimenting

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages