hasql
provides simple and reliable way to access high-availability database setups with multiple hosts.
hasql
is production-ready and is actively used inside Yandex' production environment.
With Go module support, simply add the following import
import "golang.yandex/hasql/v2"
to your code, and then go [build|run|test]
will automatically fetch the
necessary dependencies.
Otherwise, to install the hasql
package, run the following command:
$ go get -u golang.yandex/hasql/v2
hasql
operates using standard database/sql
connection pool objects. User creates *sql.DB
-compatible objects for each node of database cluster and passes them to constructor. Library keeps up to date information on state of each node by 'pinging' them periodically. User is provided with a set of interfaces to retrieve database node object suitable for required operation.
dbFoo, _ := sql.Open("pgx", "host=foo")
dbBar, _ := sql.Open("pgx", "host=bar")
discoverer := NewStaticNodeDiscoverer(
NewNode("foo", dbFoo),
NewNode("bar", dbBar),
)
cl, err := hasql.NewCluster(discoverer, hasql.PostgreSQLChecker)
if err != nil { ... }
node := cl.Primary()
if node == nil {
err := cl.Err() // most recent errors for all nodes in the cluster
}
fmt.Printf("got node %s\n", node)
ctx, cancel := context.WithTimeout(context.Background(), time.Second)
defer cancel()
if err = node.DB().PingContext(ctx); err != nil { ... }
hasql
does not configure provided connection pools in any way. It is user's job to set them up properly. Library does handle their lifetime though - pools are closed when Cluster
object is closed.
Cluster is a set of database/sql
-compatible nodes that tracks their lifespan and provides access to each individual nodes.
Node is a single database instance in high-availability cluster.
Node discoverer provides nodes objects to cluster. This abstraction allows user to dynamically change set of cluster nodes, for example collect nodes list via Service Discovery (etcd, Consul).
Node checker collects information about current state of individual node in cluster, such as: cluster role, network latency, replication lag, etc.
Node picker picks node from cluster by given criterion using predefined algorithm: random, round-robin, lowest latency, etc.
Alive primary|Alive standby|Any alive node or none otherwise
node := c.Node(hasql.Alive)
Alive primary or none otherwise
node := c.Node(hasql.Primary)
Alive standby or none otherwise
node := c.Node(hasql.Standby)
Alive primary|Alive standby or none otherwise
node := c.Node(hasql.PreferPrimary)
Alive standby|Alive primary or none otherwise
node := c.Node(hasql.PreferStandby)
When user asks Cluster
object for a node a random one from a list of suitable nodes is returned. User can override this behavior by providing a custom node picker.
Library provides a couple of predefined pickers. For example if user wants 'closest' node (with lowest latency) LatencyNodePicker
should be used.
cl, err := hasql.NewCluster(
hasql.NewStaticNodeDiscoverer(hasql.NewNode("foo", dbFoo), hasql.NewNode("bar", dbBar)),
hasql.PostgreSQLChecker,
hasql.WithNodePicker(new(hasql.LatencyNodePicker[*sql.DB])),
)
Since library requires Querier
interface, which describes a subset of database/sql.DB
methods, it supports any database that has a database/sql
driver. All it requires is a database-specific checker function that can provide node state info.
Check out node_checker.go
file for more information.
Node's state is transient at any given time. If Primary()
returns a node it does not mean that node is still primary when you execute statement on it. All it means is that it was primary when it was last checked. Nodes can change their state at a whim or even go offline and hasql
can't control it in any manner.
This is one of the reasons why nodes do not expose their perceived state to user.
You can add instrumentation via Tracer
object similar to httptrace in standard library.