Subscribe

UiPath AI Fabric

UiPath AI Fabric

1. Provision a Machine

This section details the steps that need to be taken when provisioning a machine.

📘

Note

  • AI Fabric installation is currently supported on a fresh, dedicated machine. A shared machine with unknown polices/restriction may need additional workaround and is not officially supported.
  • We recommend use of VM instances provided by popular cloud providers for AI Fabric installation. For setups that require GPU, only NVIDIA GPUs are supported. These include but are not limited to: Standard_NC12 on Azure, n1-standard-16 on GCP, p3.2xlarge on AWS.

Boot Disk

The boot disk for the machine must be at least 200 GB for normal installation and 500 GB for air-gapped installation, otherwise the installer fails.

🚧

When provisioning a VM on a cloud subscription, there is often a default boot / primary disk size. For example, the default for Azure is 30GB. Remember to expand this disk.

❗️

Airgapped

Ideally for airgapped installation the 500 GB should be mounted under the /var directory. If you can't change the size of /var due to internal policy we would need the 500 GB to be shared as follow:

  • /var -> 45GB (or as big as you can)
  • /var/lib/etcd -> 5GB
  • /var/lib/docker -> 250 GB
  • /var/lib/kubelet -> 200 GB

Data Disk

The external data disk must be attached but not formatted. It is very important the disk is unformatted (raw). If formatted, the installer fails and cannot be recovered. In this case, you would need to re-provision a new machine and attach new disks. Moreover, the secondary storage used for AI Fabric must be a disk (not a partition).

Network Configuration

🚧

Make sure you allow access to these ports from where you will access AI Fabric

8800, 31443, 31390, 443

Make sure you have inbound rules to allow access to the ports above from the network where you will access AI Fabric.

🚧

Make sure you allow inbound and outbound rules to the public IP from the AIF machine itself

6443

This port is used to access the kubernetes API, only the IP of the machine itself needs inbound and outbound access (localhost is not used by the application but rather the IP itself).

If you want to use Document Understanding models make sure that the machine can access https://du-metering.uipath.com. This is not applicable for airgapped install.

Checking Unformatted Disk

As noted above, sufficiently large external data disks that are unformatted are needed for the installation. Run the following command in a bash prompt on this Linux machine to check that the external disk is indeed unformatted:

file -sL /dev/sd*

The output looks like this:

<user>@<hostname>:~$ file -sL /dev/sd*
/dev/sda:   DOS/MBR boot sector, extended partition table (last)
/dev/sda1:  Linux rev 1.0 ext4 filesystem data, UUID=01772985-75fa-4dd1-b12c-66526c1fb0cd, volume name "cloudimg-rootfs" (needs journal recovery) (extents) (64bit) (large files) (huge files)
/dev/sda14: data
/dev/sda15: DOS/MBR boot sector, code offset 0x58+2, OEM-ID "mkfs.fat", Media descriptor 0xf8, sectors/track 32, heads 64, sectors 217088 (volumes > 32 MB), FAT (32 bit), sectors/FAT 1670, reserved 0x1, serial number 0xad0c5d13, label: "UEFI       "
/dev/sdb:   data

Notice that the value of /dev/sdb is just data which means this disk is unformatted. You can similarly check the size of the disks in your system with the following command:

lsblk --all --list --output KNAME,SIZE,TYPE,PKNAME,NAME

The output looks like this:

KNAME   SIZE TYPE PKNAME NAME
loop0 126.4M loop        loop0
loop1  55.3M loop        loop1
loop2 126.9M loop        loop2
loop3   9.5M loop        loop3
loop4  29.9M loop        loop4
loop5        loop        loop5
loop6  30.3M loop        loop6
loop7        loop        loop7
sda     100G disk        sda
sda1   99.9G part sda    sda1
sda14     4M part sda    sda14
sda15   106M part sda    sda15
sdb     200G disk        sdb

The output above shows two disks: sda of size 100GB (primary disk) and sdb of size 200GB (the external disk). Note that sdb is of TYPE disk. The step 4. Run the AI Fabric Infrastructure Installer will look for unformatted empty disks. If you have unformatted space in a device of TYPE equal to partition, the installer will not find the right device.

Checking Connectivity

From the AI Fabric machine, i.e. the machine where AI Fabric will be installed:
Check that you can connect to your Orchestrator by executing the following command:

telnet <Orchestrator Fully-Qualified-Domain-Name> 443

Check that you can connect to your SQL database by executing

telnet <SQL-database-ip> 1433

Note 1433 is the default port for SQL server, if your database is on a different port, use that port.

Requirements Before Proceeding

Requirement

Notes

Working Orchestrator 20.4 (or higher) Installation

Reference here

SQL Server 2008 R2 or above

Reference here

SQL Server Authentication must be enabled

Contact your SQL Server admin.

SQL Credentials that allow for database and role creation.

Contact your SQL Server admin.

Have a compatible browser

Reference here

Boot Disk needs to be at least 200GB (500GB if airgapped install)

Reference here

Secondary Disk(s) of at least 250GB in aggregate need to be unformatted

Reference here

Connect to Orchestrator from AI Fabric machine. Must be able to connect via domain name.

telnet <Orchestrator-Domain-Name> <port> from the AI Fabric machine must work. Reference here

Connect to Database from AI Fabric machine.

telnet <SQL-Server-IP> <port> from the AI Fabric machine must work. Reference here

Connect from Robot/Studio machines to AI Fabric.

telnet <aif-linux-machine> 31390 from robot/studio machines must work. That is, robot/studio machine will be clients to the linux machine on port 31390.

Connect to Endpoints needed by installer from AI Fabric machine (not applicable for airgapped)

AIFabric machine must not have blocked outbound access to these endpoints

Domain Certificate (for AI Fabric machine) from a trusted CA authority.

Reference here

AI Fabric license file.

Reference here

Updated about a month ago



1. Provision a Machine


Suggested Edits are limited on API Reference Pages

You can only suggest edits to Markdown body content, but not to the API spec.