# Upgrading From Version 1

Version 2 of ArchitectureKit builds on version 2 of the **[client SDK for Go](/docs/eventsourcingdb/sdks/go)**, whose types the kit takes and hands out, such as `*eventsourcingdb.Client` and `eventsourcingdb.Event`, so the two move together. Version 1 of the kit is no longer maintained.

Upgrading takes the following changes.

## Changing the Module Path

The module path ends with `/v2`. Import `github.com/thenativeweb/architecturekit-golang/v2/architecturekit` and the other packages below `github.com/thenativeweb/architecturekit-golang/v2`, together with `github.com/thenativeweb/eventsourcingdb-client-golang/v2/eventsourcingdb`:

```go
import (
  "github.com/thenativeweb/architecturekit-golang/v2/architecturekit"
  "github.com/thenativeweb/architecturekit-golang/v2/architecturekit/httpapi"
  "github.com/thenativeweb/eventsourcingdb-client-golang/v2/eventsourcingdb"
)
```

The same goes for `query`, `architecturekittest`, and `dbtest`, such as `github.com/thenativeweb/architecturekit-golang/v2/architecturekit/architecturekittest/dbtest`. For what changes in the client SDK, see its **[notes on upgrading](/docs/eventsourcingdb/sdks/go#upgrading-from-version-1)**.

## Sorting Failures by Their Code

A failure of the database is sorted by its code (see **[Handling Errors](/docs/architecturekit/handling-errors)**). A database that offers the v2 API reports the errors of a request with the status `200`, so the status no longer tells them apart. Almost every failure keeps its category, apart from these:

- A store that has reached the capacity of its license (`capacity-exceeded`) belongs to `ErrTransient` now rather than to `ErrConflict`, so a store that decides again on conflicts no longer does so in vain (see **[Guarding Against Concurrent Changes](/docs/architecturekit/using-preconditions#guarding-against-concurrent-changes)**), and the `httpapi` package answers it with `503 Service Unavailable` rather than `409 Conflict` (see **[Mapping Errors to Status Codes](/docs/architecturekit/mapping-errors-to-status-codes)**).
- A request that did not arrive in time (`408`) belongs to `ErrTransient` now rather than to `ErrPermanent`.

## Reporting Unknown Outcomes

A write that was cut off after the client had connected is reported as `ErrOutcomeUnknown`, rather than as `ErrTransient`. Version 1 traced the request itself, to tell one that never left completely. Version 2 follows the client, which applies the same rule in every SDK of EventSourcingDB, and reports such a write with `eventsourcingdb.ErrOutcomeUnknown` (see **[Handling Errors](/docs/architecturekit/handling-errors)**).

## Finding the Errors of the Client

`errors.Is` finds `eventsourcingdb.ErrConnectionStalled` rather than `eventsourcingdb.ErrHeartbeatTimeout`, and the `*eventsourcingdb.DBAPIError` that `errors.As` finds has a `Status`, a `Code`, and a `Description`, rather than a `StatusCode` and a `Reason`:

```go
var dbAPIError *eventsourcingdb.DBAPIError
if errors.As(err, &dbAPIError) {
  log.Println(dbAPIError.Status, dbAPIError.Code, dbAPIError.Description)
}
```

## Handling Streams That End

A stream that ends counts as broken off. The `Err` of a `Reconnect` and of a `ProjectionStatus` used to be `nil` then, and is an error of the category `ErrTransient` now, which says that the stream broke off (see **[Running Projections](/docs/architecturekit/projections#running-projections)**).