Endpoints

An endpoint is a stored connection. Stream keeps the address and the credentials
in one place, gives it a name, and every task afterwards refers to it by that name.
Nobody retypes a password into a task, and when a server moves you change it once.

Open Endpoints from the Stream menu.

Each row is one connection. Direction is the column to read first: it tells
you whether the connection is somewhere Stream reads from, somewhere it writes to,
or both.

Column	What it tells you
Name	What tasks will call this connection
Type	The kind of system at the other end
System	The specific product: AF+PI, mssql, file
Host	The server or folder it points at
Direction	Source, Destination, or Both
Status	Whether the last check reached it
Last Checked	When that check happened

One rule that shapes everything

Only a PI System can be read from. Every other endpoint type is write-only.

Stream takes process data out of your PI System and puts it somewhere useful, so
every job has a PI System on the input side. The only question is where the results
go.

Set up your PI System connection first. You will need it before you can build a
pattern or a task.

Add an endpoint

Select Add Endpoint. Stream asks which kind you want before it asks for any
details.

The types are grouped, and the group heading tells you the direction: the
Source group can be read from, every other group is write-only. Pick a card and
select Continue ›, or double-click the card.

Some cards are visible but cannot be selected yet. They show what Stream is being
built towards; the ones you can configure today are listed below. If Continue ›
stays unavailable after you pick a card, that type is not ready yet.

What you can connect to today

Each type has its own page: the fields it takes, and the traps particular to it.

Type	Group	Direction	Page
AVEVA AF / PI	Source	Read and write	AVEVA AF / PI
File system	Files & object stores	Write	File system
Databricks	Files & object stores	Write	Databricks
Snowflake	Files & object stores	Write	Snowflake
Microsoft SQL Server	Databases	Write	Microsoft SQL Server
Oracle	Databases	Write	Oracle
AVEVA CDS	Industrial cloud	Coming soon	AVEVA CDS
Cognite CDF	Industrial cloud	Coming soon	Cognite CDF

Some cards are visible but cannot be selected yet. They show what Stream is being
built towards; if Continue › stays unavailable after you pick a card, that type
is not ready.

AVEVA CDS and Cognite CDF sit between the two. Their forms work and an
endpoint saves, but no task can write to one yet, so they are marked coming
soon above.

Two of them do more than connect

Most destinations write and are finished. Databricks and Snowflake land
Parquet files and can then load those files into a table: two steps, and the
second one is set per task rather than on the connection. See
Loading a Lake Table.

Check that a connection works

There are two ways, and they answer slightly different questions.

From the row. Each row has a refresh control in the Actions column. It tests
the endpoint exactly as the service uses it: the stored configuration, stored
credentials, and updates Status and Last Checked.

From the form. The Add and Edit dialogs have a Test Connection button. This
tests what is on screen, before you commit it, which is the one that saves you
time: a button that only works after saving trains people to save broken endpoints.

Reading the result

The result is not a yes or no. It is a list of graded checks, in the order the
failures actually happen, each with its own outcome and a line written for a person:

Type	Checks, in order
File system	directory exists → writable (a probe file is written and removed)
Microsoft SQL Server	connection → SELECT @@VERSION
Databricks	credentials → volume write and delete → SQL warehouse
Snowflake	private key → credentials → stage LIST → warehouse

The list stops at the first failure. A volume check after a failed token would be
a second red row for one problem. What you see is a prefix, not a full matrix, and the
last row is the one to read.

The grading separates three independent failures. "Connection failed" on its own
would leave you to guess which of them you have.

The four outcomes

Outcome	What it means
Passed	Everything Stream checked worked
Failed	A specific check failed; read its detail line
Not supported	This endpoint type has no test yet. Not a verdict either way
Not tested	Nothing answered at all

The last two are worth separating from a failure. Not supported means Stream
has no test for that type: painting it green would claim something nobody
verified, and red would accuse a configuration that may be perfectly correct.
Not tested means the call never got there: the Stream service is down, or the
request was refused. The endpoint was not tested, so nothing has been learnt about
it, and re-reading its fields is wasted effort.

Status on the list

Status	Meaning
Connected	Stream reached it on the last check
Idle	Not checked recently: normal for a connection nothing is currently using
Error	The last check failed

Idle is not a problem. It means no task has needed this connection lately. Test
it directly if you want to confirm it before building on it.

Editing an endpoint does not expose its secret

A stored password or key is never sent back to the browser: the field shows empty
with "unchanged: type to replace". Leave it alone and the stored value is kept,
on both Test Connection and Save. You only ever type a secret to change it.

If a check fails, the detail line beside it is the first thing to read; it carries
the provider's own message. The Logs page has the rest.

What's Next

Set up the PI System connection first: nothing else can be built until it exists.
Continue to AVEVA AF / PI →.

Related

• Getting Started: the whole sequence end to end
• Task Output: choosing the destination and format for a task
• Loading a Lake Table: what Databricks and Snowflake do after landing files
• Data Sources: the drivers behind IOTA Vue's own data connections