This is required to allow new code owned by ScyllaDB.

Signed-off-by: Piotr Jastrzebski <[email protected]>
Signed-off-by: Piotr Jastrzebski <[email protected]>
Signed-off-by: Piotr Jastrzebski <[email protected]>

Part of Connection initialization will be obtaining
info about shards in the host so Connection needs
a reference to the host to be able to set up this info.

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>
Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

This field will track the details of shards available at the host.
It will be set during connection initialization.
Multiple such initializations can occure at the same time but
each should be setting the same number of shards.

Signed-off-by: Piotr Jastrzebski <[email protected]>

This field will be set during initialization and will
describe the shard in the host that a connection is associated to.

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

This field will be used for paging queries to fetch all
pages from the same host if possible.

Signed-off-by: Piotr Jastrzebski <[email protected]>

This policy will try to fetch all pages from
the same host if possible.

Signed-off-by: Piotr Jastrzebski <[email protected]>
Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>
Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

HostConnectionPool.cleanupIdleConnections is called from
a recurring task. It can race with the initialization
of the pool leading to NPE and other concurrency problems.

Signed-off-by: Piotr Jastrzebski <[email protected]>

By default the driver has the ability to coalesce schem refresh queries
to try and reduce the number of queries sent. In case a view and a table
refresh is needed a refresh for the keyspace would be generated.

While in most cases thisoptimization is helpfull - in case there are
100s of tables in a single keyspace this will cause fetching all
keyspace information which will cause extra load on the system (in
scylla's case a single shard).

Add the ability to disable this by reusing setting of
maxPendingRefreshSchmaRequests value. If its 1 disable coalescing.

Signed-off-by: Shlomi Livne <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Make sure groupId is set everywhere to com.scylladb and
artifactId has scylla- prefix instead of cassandra- prefix.

Signed-off-by: Piotr Jastrzebski <[email protected]>

Those metrics expose the information the driver has
about number of shards each node has.

New metric is called shard-awareness-info and its
value is a Map<com.datastax.driver.core.Host, Integer>.

For each host (which represents a node in a cluster) it presents
either the number of shards the host has or null if the node
does not provide sharding information.

Signed-off-by: Piotr Jastrzebski <[email protected]>

Fixes scylladb#14

Signed-off-by: Piotr Jastrzebski <[email protected]>

Signed-off-by: Piotr Jastrzebski <[email protected]>

Previously the pool wasn't available for users before we have created
requested number of connections to each shard. This was realized by
not making a future returned from initAsync ready.

Some users were reporting the initialization taking
too long and timing out.

This commit changes the way initAsync works. From now on:
1. initAsync will try to open 2 * |number of shards| connections
2. Then it will close connections that are excessive for each shard
3. For shards that didn't get enough connections, connection tasks
   will be scheduled to create missing connections.

Also we are limiting the number of connections each ConnectionTask
holds to no more than the number of shards.

Signed-off-by: Piotr Jastrzebski <[email protected]>

if connections for a target shard are still being
initialized.

Signed-off-by: Piotr Jastrzebski <[email protected]>

Previously each ConnectionTask was keeping its own excessive connections
(connections that are not to the shard it wants).

This patch changes the behaviour so that all ConnectionTasks
share excessive connections. This way we not only better limit
number of excessive connections but also will quicker find
the right connection for each task (some other task may open
a connection for it).

Signed-off-by: Piotr Jastrzebski <[email protected]>

HostConnectionPool: finish initializing after getting first connections

…Fixed force push)