Load balancing

Load balancing is turned on by default for any distributed index using mirroring. By default queries are distributed randomly among the mirrors. To change this behaviour you can use ha_strategy.

ha_strategy

ha_strategy = {random|nodeads|noerrors|roundrobin}

Agent mirror selection strategy for load balancing. Optional, default is random.

The strategy used for mirror selection, or in other words, choosing a specific agent mirror in a distributed index. Essentially, this directive controls how exactly master does the load balancing between the configured mirror agent nodes. The following strategies are implemented:

Simple random balancing

The default balancing mode. Simple linear random distribution among the mirrors. That is, equal selection probability are assigned to every mirror. Kind of similar to round-robin (RR), but unlike RR, does not impose a strict selection order.

  • Example
Example
📋
ha_strategy = random

Adaptive randomized balancing

The default simple random strategy does not take mirror status, error rate and, most importantly, actual response latencies into account. So to accommodate for heterogeneous clusters and/or temporary spikes in agent node load, we have a group of balancing strategies that dynamically adjusts the probabilities based on the actual query latencies observed by the master.

The adaptive strategies based on latency-weighted probabilities basically work as follows:

  • latency stats are accumulated in blocks of ha_period_karma seconds;
  • once per karma period latency-weighted probabilities get recomputed;
  • once per request (including ping requests) "dead or alive" flag is adjusted.

Currently, we begin with equal probabilities (or percentages, for brevity), and on every step, scale them by the inverse of the latencies observed during the last "karma" period, and then renormalize them. For example, if during the first 60 seconds after the master startup 4 mirrors had latencies of 10, 5, 30, and 3 msec/query respectively, the first adjustment step would go as follow:

  • initial percentages: 0.25, 0.25, 0.25, 0.25;
  • observed latencies: 10 ms, 5 ms, 30 ms, 3 ms;
  • inverse latencies: 0.1, 0.2, 0.0333, 0.333;
  • scaled percentages: 0.025, 0.05, 0.008333, 0.0833;
  • renormalized percentages: 0.15, 0.30, 0.05, 0.50.

Meaning that the 1st mirror would have a 15% chance of being chosen during the next karma period, the 2nd one a 30% chance, the 3rd one (slowest at 30 ms) only a 5% chance, and the 4th and the fastest one (at 3 ms) a 50% chance. Then, after that period, the second adjustment step would update those chances again, and so on.

The rationale here is, once the observed latencies stabilize, the latency weighted probabilities stabilize as well. So all these adjustment iterations are supposed to converge at a point where the average latencies are (roughly) equal over all mirrors.

nodeads

Latency-weighted probabilities, but dead mirrors are excluded from the selection. "Dead" mirror is defined as a mirror that resulted in multiple hard errors (eg. network failure, or no answer, etc) in a row.

  • Example
Example
📋
ha_strategy = nodeads

noerrors

Latency-weighted probabilities, but mirrors with worse errors/success ratio are excluded from the selection.

  • Example
Example
📋
ha_strategy = noerrors

Round-robin balancing

Simple round-robin selection, that is, selecting the 1st mirror in the list, then the 2nd one, then the 3rd one, etc, and then repeating the process once the last mirror in the list is reached. Unlike with the randomized strategies, RR imposes a strict querying order (1, 2, 3, ..., N-1, N, 1, 2, 3, ... and so on) and guarantees that no two subsequent queries will be sent to the same mirror.

  • Example
Example
📋
ha_strategy = roundrobin

Instance-wide options

ha_period_karma

ha_period_karma = 2m

ha_period_karma defines agent mirror statistics window size, in seconds (or time suffixed). Optional, default is 60.

For a distributed index with agent mirrors in it, server tracks several different per-mirror counters. These counters are then used for failover and balancing. (Server picks the best mirror to use based on the counters.) Counters are accumulated in blocks of ha_period_karma seconds.

After beginning a new block, master may still use the accumulated values from the previous one, until the new one is half full. Thus, any previous history stops affecting the mirror choice after 1.5 times ha_period_karma seconds at most.

Despite that at most 2 blocks are used for mirror selection, up to 15 last blocks are actually stored, for instrumentation purposes. They can be inspected using SHOW AGENT STATUS statement.

ha_ping_interval

ha_ping_interval = 3s

ha_ping_interval defines interval between agent mirror pings, in milliseconds (or time suffixed). Optional, default is 1000.

For a distributed index with agent mirrors in it, server sends all mirrors a ping command during the idle periods. This is to track the current agent status (alive or dead, network roundtrip, etc). The interval between such pings is defined by this directive.

To disable pings, set ha_ping_interval to 0.

Setting up replication

Write transaction (any result of INSERT, REPLACE, DELETE, TRUNCATE, UPDATE, COMMIT) can be replicated to other cluster nodes before the transaction is fully applied on the current node. Currently replication is supported for percolate and rt indexes in Linux an MacOS. Manticore Search packages for Windows do not provide replication support.

Manticore's replication is based on Galera library and features the following:

  • true multi-master - read and write to any node at any time
  • virtually synchronous replication - no slave lag, no data is lost after a node crash
  • hot standby - no downtime during failover (since there is no failover)
  • tightly coupled - all the nodes hold the same state. No diverged data between nodes allowed
  • automatic node provisioning - no need to manually back up the database and restore it on a new node
  • easy to use and deploy
  • detection and automatic eviction of unreliable nodes
  • certification based replication

To use replication in Manticore Search:

  • data_dir option should be set in section "searchd" of the configuration file. Replication is not supported in the plain mode
  • there should be either:
    • a listen directive specified (without specifying a protocol) containing an IP address accessible by other nodes
    • or node_address with an accessible IP address
  • optionally you can set unique values for server_id on each cluster node. If no value is set, the node will try to use the MAC address (or a random number if that fails) to generate the server_id.

If there is no replication listen directive set Manticore will use the first two free ports in the range of 200 ports after the default protocol listening port for each created cluster. To set replication ports manually the listen directive (of replication type) port range should be defined and the address/port range pairs should not intersect between different nodes on the same server. As a rule of thumb, the port range should specify no less than two ports per cluster.

Replication cluster

Replication cluster is a set of nodes among which a write transaction gets replicated. Replication is configured on per-index basis, meaning that one index can be assigned to only one cluster. There is no restriction on how many indexes a cluster can have. All transactions such as INSERT, REPLACE, DELETE, TRUNCATE in any percolate or real-time index belonging to a cluster are replicated to all the other nodes in the cluster. Replication is multi-master, so writes to any particular node or to multiple nodes simultaneously work equally well.

In most cases you create cluster with CREATE CLUSTER <cluster name> and join cluster with JOIN CLUSTER <cluster name> at 'host:port', but in rare cases you may want to fine-tune the behaviour of CREATE/JOIN CLUSTER. The options are:

name

Specifies cluster name. Should be unique.

path

Data directory for a write-set cache replication and incoming indexes from other nodes. Should be unique among the other clusters in the node. Default is data_dir.

nodes

List of address:port pairs for all the nodes in the cluster (comma separated). Node's API interface should be used for this option. It can contain the current node's address too. This list is used to join a node to the cluster and rejoin it after restart.

options

Other options that are passed over directly to Galera replication plugin as described here Galera Documentation Parameters

Write statements

For SQL interface all write statements such as INSERT, REPLACE, DELETE, TRUNCATE, UPDATE that change the content of a cluster's index should use cluster_name:index_name expression in place of an index name to make sure the change is propagated to all replicas in the cluster. An error will be triggered otherwise.

All write statements for HTTP interface to a cluster's index should set cluster property along with index name. An error will be triggered otherwise.

Auto ID generated for an index in a cluster should be valid as soon as server_id is not misconfigured.

  • SQL
  • HTTP
  • PHP
  • Python
  • Javascript
  • Java
📋
INSERT INTO posts:weekly_index VALUES ( 'iphone case' )
TRUNCATE RTINDEX click_query:weekly_index
UPDATE INTO posts:rt_tags SET tags=(101, 302, 304) WHERE MATCH ('use') AND id IN (1,101,201)
DELETE FROM clicks:rt WHERE MATCH ('dumy') AND gid>206

Read statements

Read statements such as SELECT, CALL PQ, DESCRIBE can use either regular index names not prepended with a cluster name or cluster_name:index_name. In this case cluster_name is just ignored.

In HTTP endpoint json/search you can specify cluster property if you like, but can also omit it.

  • SQL
  • HTTP
📋
SELECT * FROM weekly_index
CALL PQ('posts:weekly_index', 'document is here')

Cluster parameters

Replication plugin options can be changed using SET statement (see the example).

See Galera Documentation Parameters for a list of available options.

  • SQL
  • HTTP
📋
SET CLUSTER click_query GLOBAL 'pc.bootstrap' = 1

Cluster with diverged nodes

Sometimes replicated nodes can diverge from each other. The state of all the nodes might turn into non-primary due to a network split between nodes, a cluster crash, or if the replication plugin hits an exception when determining the primary component. Then it's necessary to select a node and promote it to the primary component.

To determine which node needs to be a reference, compare the last_committed cluster status variable value on all nodes. If all the servers are already running there's no need to start the cluster again. You just need to promote the most advanced node to the primary component with SET statement (see the example).

All other nodes will reconnect to the node and resync their data based on this node.

  • SQL
  • HTTP
📋
SET CLUSTER posts GLOBAL 'pc.bootstrap' = 1

Replication and cluster

To use replication define one listen port for SphinxAPI protocol and one listen for replication address and port range in the config. Define data_dir folder for incoming indexes.

  • ini
ini
📋
searchd {
  listen   = 9312
  listen   = 192.168.1.101:9360-9370:replication
  data_dir = /var/lib/manticore/
  ...
 }

Create a cluster at the server that has local indexes that need to be replicated

  • SQL
  • HTTP
  • PHP
  • Python
  • Javascript
  • Java
📋
CREATE CLUSTER posts

Add these local indexes to cluster

  • SQL
  • HTTP
  • PHP
  • Python
  • Javascript
  • Java
📋
ALTER CLUSTER posts ADD pq_title
ALTER CLUSTER posts ADD pq_clicks

All other nodes that want replica of cluster's indexes should join cluster as

  • SQL
  • HTTP
  • PHP
  • Python
  • Javascript
  • Java
📋
JOIN CLUSTER posts AT '192.168.1.101:9312'

When running queries for SQL prepend the index name with the cluster name posts: or use cluster property for HTTP request object.

  • SQL
  • HTTP
  • PHP
  • Python
  • Javascript
  • Java
📋
INSERT INTO posts:pq_title VALUES ( 3, 'test me' )

Now all such queries that modify indexes in the cluster are replicated to all nodes in the cluster.

Creating a replication cluster

To create a replication cluster you should set at least its name.

In case of a single cluster or if the cluster you are creating is the first one path option may be omitted, in this case data_dir option will be used as the cluster path. For all subsequent clusters you need to specify path and this path should be available. nodes option may be also set to enumerate all the nodes in the cluster.

  • SQL
  • HTTP
  • PHP
  • Python
  • javascript
  • Java
📋
CREATE CLUSTER posts
CREATE CLUSTER click_query '/var/data/click_query/' as path
CREATE CLUSTER click_query '/var/data/click_query/' as path, 'clicks_mirror1:9312,clicks_mirror2:9312,clicks_mirror3:9312' as nodes

If a cluster is created without the nodes option, the first node that gets joined to the cluster will be saved as nodes.