layout | title | permalink | redirect_from | ||
---|---|---|---|---|---|
post |
GETTING STARTED |
/docs/getting-started |
|
AIStore runs on a single Linux or Mac machine. Bare-metal Kubernetes and GCP/GKE Cloud-based deployments are also supported. There are numerous other options.
Generally, when deciding how to deploy a system like AIS with so many possibilities to choose from, a good place to start would be answering the following two fundamental questions:
- what's the dataset size, or sizes?
- what hardware will I use?
For datasets, say, below 50TB a single host may suffice and should, therefore, be considered a viable option. On the other hand, Cloud deployment option may sound attractive for its ubiquitous convenience and for not thinking about the hardwares and the sizes - at least, not right away.
Note as well that you can always start small: a single-host deployment, a 3-node cluster in the Cloud or on premises, etc. AIStore supports a number of options to inter-connect existing clusters - the capability called unified global namespace - or migrate existing datasets (on demand or via supported storage services). For introductions and further pointers, please refer to the AIStore Overview.
AIStore runs on commodity Linux machines with no special hardware requirements whatsoever.
It is expected that within a given cluster, all AIS target machines are identical, hardware-wise.
- Linux (with
gcc
,sysstat
andattr
packages, and kernel 4.15+) or MacOS - Go 1.17 or later
- Extended attributes (
xattrs
- see below) - Optionally, Amazon (AWS) or Google Cloud Platform (GCP) account(s)
Depending on your Linux distribution, you may or may not have gcc
, sysstat
, and/or attr
packages.
The capability called extended attributes, or xattrs
, is a long time POSIX legacy supported by all mainstream filesystems with no exceptions. Unfortunately, xattrs
may not always be enabled in the Linux kernel configurations - the fact that can be easily found out by running setfattr
command.
If disabled, please make sure to enable xattrs in your Linux kernel configuration.
MacOS/Darwin is also supported, albeit for development only. Certain capabilities related to querying the state-and-status of local hardware resources (memory, CPU, disks) may be missing, which is why we strongly recommend Linux for production deployments.
AIS comes with its own build system that we use in a variety of development and production environments. The very first make
command to run would be:
$ make help
The output - truncated below for the sake of brevity - includes supported subcommands, environment variables, and usage examples:
Useful commands:
aisfs Build 'aisfs' binary
aisloader Build 'aisloader' binary
all Build all main binaries
authn Build 'authn' binary
ci Run CI related checkers and linters (requires BUCKET variable to be set)
clean-client-bindings Remove all generated client binding files
clean Remove all AIS related files and binaries
cli-autocompletions Add CLI autocompletions
cli Build CLI ('ais' binary)
deploy Build 'aisnode' and deploy the specified numbers of local AIS proxies and targets
deploy-docker Deploy AIS cluster inside the dockers
...
...
Examples:
# Deploy cluster locally
$ make deploy
# Stop locally deployed cluster and cleanup all cluster-related data and bucket metadata (but not cluster map)
$ make kill clean
# Stop and then deploy (non-interactively) cluster consisting of 7 targets (4 mountpaths each) and 2 proxies; build `aisnode` executable with the support for GCP and AWS backends
$ make kill deploy <<< $'7\n4\n4\ny\ny\nn\nn\n'
# Shutdown and then (non-interactively) generate local configs and deploy a cluster consisting of 7 targets (4 mountpaths each) and 2 proxies; build executable with the support for GCP and AWS backends
$ make kill deploy <<< $'7\n4\n4\ny\ny\nn\nn\n'
# Restart a cluster of 7 targets (4 mountpaths each) and 2 proxies; utilize previously generated (pre-shutdown) local configurations
$ make restart <<< $'7\n4\n4\ny\ny\nn\nn\n'
# Redeploy (4 targets, 1 proxy) cluster; build executable for debug without any backend-supporting libraries; use RUN_ARGS to pass an additional command-line option ('-override_backends=true') to each running node
$ RUN_ARGS=-override_backends MODE=debug make kill deploy <<< $'4\n1\n4\nn\nn\nn\nn\n'
# Same as above, but additionally run all 4 targets in a standby mode
$ RUN_ARGS='-override_backends -standby' MODE=debug make kill deploy <<< $'4\n1\n4\nn\nn\nn\nn\n'
...
...
For any Kubernetes deployments (including, of course, production deployments) please use a separate and dedicated AIS-K8s GitHub repository. The repo contains Helm Charts and detailed Playbooks that cover a variety of use cases and configurations.
In particular, AIS-K8s GitHub repository provides a single-line command to deploy Kubernetes cluster and the underlying infrastructure with AIStore cluster running inside (see below). The only requirement is having a few dependencies preinstalled (in particular, helm
) and a Cloud account.
The following GIF illustrates steps to deploy AIS on Google Cloud Platform (GCP):
Finally, the repository hosts Kubernetes Operator project that will eventually replace Helm charts and will become the main deployment, lifecycle, and operation management "vehicle" for AIStore.
This option has the unmatched convenience of requiring an absolute minimum time and resources - please see this README for details.
You could use the instruction below for a quick evaluation, experimenting with features, first-time usage - and, of course, for development. Local AIStore playground is not intended for production clusters and is not meant to provide optimal performance.
The following video gives a quick intro to AIStore, along with a brief demo of the local playground and development environment.
{% include youtubePlayer.html id="ANshjHphqfI" %}
Assuming that Go toolchain is already installed, the steps to deploy AIS locally on a single development machine are:
$ cd $GOPATH/src/github.com/NVIDIA
$ git clone https://github.com/NVIDIA/aistore.git
$ cd aistore
# optionally, run `make mod-tidy` to preload dependencies
$ ./deploy/scripts/clean_deploy.sh
$ ais show cluster
where:
clean_deploy.sh
with no arguments builds AIStore binaries (such asaisnode
andais
CLI) and then deploys a local cluster with 5 proxies and 5 targets. Examples:
# Deploy 7 targets and 1 proxy:
$ clean_deploy.sh --proxy-cnt 1 --target-cnt 7
# Same as above, plus built-in support for GCP (cloud storage):
$ clean_deploy.sh --proxy-cnt 1 --target-cnt 7 --gcp
For more options and detailed description, run make help
and see: clean_deploy.sh
.
You can also run make deploy
in the root directory of the repository to deploy a cluster:
$ make deploy
Enter number of storage targets:
10
Enter number of proxies (gateways):
3
Number of local cache directories (enter 0 to use preconfigured filesystems):
2
Select backend providers:
Amazon S3: (y/n) ?
n
Google Cloud Storage: (y/n) ?
n
Azure: (y/n) ?
n
HDFS: (y/n) ?
n
Would you like to create loopback mount points: (y/n) ?
n
Building aisnode: version=df24df77 providers=
Notice the "Cloud" prompt above, and the fact that access to 3rd party Cloud storage is a deployment-time option.
make kill
will terminate local AIStore if it's already running.
For more development options and tools, please refer to the development docs.
For development, health-checking a new deployment, or for any other (functional and performance testing) related reason you can run any/all of the included tests.
For example:
$ go test ./ais/tests -v -run=Mirror
The go test
above will create an AIS bucket, configure it as a two-way mirror, generate thousands of random objects, read them all several times, and then destroy the replicas and eventually the bucket as well.
Alternatively, if you happen to have Amazon and/or Google Cloud account, make sure to specify the corresponding (S3 or GCS) bucket name when running go test
commands.
For example, the following will download objects from your (presumably) S3 bucket and distribute them across AIStore:
$ BUCKET=aws://myS3bucket go test ./ais/tests -v -run=download
In our development and testing, we make use of Minikube and the capability, further documented here, to run Kubernetes cluster on a single development machine. There's a distinct advantage that AIStore extensions that require Kubernetes - such as Extract-Transform-Load, for example - can be developed rather efficiently.
In the end, all examples above run a bunch of local web servers that listen for plain HTTP requests. Following are quick steps for developers to engage HTTPS:
- Generate X.509 certificate:
$ openssl req -x509 -newkey rsa:4096 -keyout server.key -out server.crt -days 1080 -nodes -subj '/CN=localhost'
- Deploy cluster (4 targets, 1 gateway, 6 mountpaths, Google Cloud):
$ AIS_USE_HTTPS=true AIS_SKIP_VERIFY_CRT=true make kill deploy <<< $'4\n1\n6\nn\ny\nn\nn\nn\n'
- Run tests (both examples below list the names of buckets accessible for you in Google Cloud):
$ AIS_ENDPOINT=https://localhost:8080 AIS_SKIP_VERIFY_CRT=true BUCKET=gs://myGCPbucket go test -v -p 1 -count 1 ./ais/tests -run=ListBuckets
$ AIS_ENDPOINT=https://localhost:8080 AIS_SKIP_VERIFY_CRT=true BUCKET=tmp go test -v -p 1 -count 1 ./ais/tests -run=ListBuckets
Notice environment variables above: AIS_USE_HTTPS, AIS_ENDPOINT, and AIS_SKIP_VERIFY_CRT.
As noted, the project utilizes GNU make
to build and run things both locally and remotely (e.g., when deploying AIStore via Kubernetes. As the very first step, run make help
for help on:
- building AIS binary (called
aisnode
) deployable as both a storage target or a proxy/gateway; - building CLI, aisfs, and benchmark binaries;
In particular, the make
provides a growing number of developer-friendly commands to:
- deploy AIS cluster on your local development machine;
- run all or selected tests;
- instrument AIS binary with race detection, CPU and/or memory profiling, and more.
The following applies to all containerized deployments:
- AIS nodes always automatically detect containerization.
- If deployed as a container, each AIS node independently discovers whether its own container's memory and/or CPU resources are restricted.
- Finally, the node then abides by those restrictions.
To that end, each AIS node at startup loads and parses cgroup settings for the container and, if the number of CPUs is restricted, adjusts the number of allocated system threads for its goroutines.
This adjustment is accomplished via the Go runtime GOMAXPROCS variable. For in-depth information on CPU bandwidth control and scheduling in a multi-container environment, please refer to the CFS Bandwidth Control document.
Further, given the container's cgroup/memory limitation, each AIS node adjusts the amount of memory available for itself.
Limits on memory may affect dSort performance forcing it to "spill" the content associated with in-progress resharding into local drives. The same is true for erasure-coding that also requires memory to rebuild objects from slices, etc.
For technical details on AIS memory management, please see this readme.