3. Server Parameters

3. Server Parameters

3.1. New and Modified Server Parameters

variable auditDestination

Supported since 2.0.0

Has no effect in Percona TokuMX Community Edition. See Audit for more details.

The type of audit log that will be created. The only currently supported value is file.

variable auditFilter
Default Value:{}

Supported since 2.0.0

Has no effect in Percona TokuMX Community Edition. See Audit for more details.

A JSON object used to filter audit events. Should be a valid MongoDB query; only audit events that match this query will be logged. By default, all events are logged.

variable auditFormat

Supported since 2.0.0

Has no effect in Percona TokuMX Community Edition. See Audit for more details.

The format used for audit events. The only currently supported value is JSON.

variable auditPath

Supported since 2.0.0

Has no effect in Percona TokuMX Community Edition. See Audit for more details.

For the file auditDestination, the filename used for the audit log. By default, this is auditLog.json in the same directory as logpath if it exists, otherwise inside dbpath for mongod or inside the current working directory for mongos.

variable cacheSize
Default Value:1/2 of physical RAM, computed at startup.

Amount of memory (in bytes) used to cache documents and indexes.

Can supply values using K/M/G (i.e., ‘4G’ = 4 Gigabytes)

variable checkpointPeriod
Default Value:60 seconds.

This variable specifies the number of seconds between checkpoints.

This parameter actually controls the minimum number of seconds between the start of consecutive checkpoints. If a checkpoint takes longer than 60 seconds, the subsequent checkpoint will begin immediately upon its predecessor’s completion.

variable cleanerIterations
Default Value:5

This variable specifies how many internal nodes get processed in each cleaner thread run.

Setting this variable to 0 turns off cleaner threads.

variable cleanerPeriod
Default Value:2 seconds

This variable specifies the number of seconds between cleaner thread runs.

Setting this variable to 0 turns off cleaner threads.

variable dbpath
Default Value:/data/db

Directory where the Percona TokuMX collections (including the oplog for replication) are stored.

variable directio
Default Value:false

Use direct I/O instead of buffered I/O.

Out of memory workloads typically perform better with direct I/O.

When using direct I/O, we recommend a cacheSize of 80% physical RAM or more.

variable expireOplogDays / expireOplogHours
Default Value:14 days

Number of days’ worth of data to keep in the primary’s local.oplog.rs. This is effectively how far behind a secondary can get before it needs to do a new initial sync when it gets re-integrated into the set.

This parameter only affects replica sets.

Note

This is different behavior than MongoDB, where the oplog is a capped collection. You should change this default to the appropriate number of days or hours that you want to retain in the oplog.

variable fastUpdates
Default Value:false

Supported since 2.0.0

If this setting is true, then Fast Updates are enabled.

variable fastUpdatesIgnoreErrors
Default Value:false

Supported since 2.0.0

If a Fast Updates results in an error (e.g. incrementing a text field), the error may not be encountered until some undetermined time in the future. If this setting is true, such errors are ignored. If false, such errors are occasionally written to the error log.

variable fsRedzone
Default Value:5%

This variable controls the percentage of the file system that must be available for inserts to be allowed.

variable loadPlugin

Specified as loadPlugin=name:checksum.

Specifies that the plugin name from libname.so should be loaded as a plugin on startup, and that it will not be loaded if it does not have the proper checksum.

You can determine this checksum by running the loadPlugin command manually, as db.adminCommand({loadPlugin: "name"}). The checksum will be returned in the result object.

This option can be specified multiple times, in a config file or on the command line. The plugins available are Hot Backup and Point in Time Recovery.

variable loaderCompressTmp
Default Value:true

Whether the Bulk Loader will compress intermediate files (see tmpDir) while building indexes.

variable loaderMaxMemory
Default Value:100MB

Controls the amount of memory used by the Bulk Loader.

Can supply values using K/M/G (i.e., ‘400M’ = 400 Megabytes)

variable lockTimeout
Default Value:4000 ms

This variable controls the amount of time in milliseconds that a transaction will wait for a lock held by another transaction to be released. If the conflicting transaction does not release the lock within the lock timeout, the transaction that was waiting for the lock will get a lock timeout error.

A value of 0 disables lock waits.

Note

See Document-level Lock Conflicts for details about conflicting transactions.

variable locktreeMaxMemory
Default Value:10% of cacheSize

Maximum amount of memory (in bytes) used by the document-level locking data structures.

variable logDir
Default Value:same directory as dbpath

Directory where the Percona TokuMX transaction log (similar to the MongoDB durability journal) will go.

variable logFlushPeriod / journalCommitInterval
Default Value:100ms
Values:0ms - 300ms

How often to fsync the recovery log (like journalCommitInterval in basic MongoDB).

Unlike original MongoDB, a value of 0 means commit every operation.

Warning

Committing every operation can be expensive.

variable pluginsDir
Default Value:installdir/lib64/plugins

The directory searched for plugins by loadPlugin.

Usually should not be modified.

variable rsMaintenance
Default Value:false

Supported since 2.0.0

Starts the server in maintenance mode

Has no effect if the server is not a replica set member.

Used for Point in Time Recovery.

variable tmpDir
Default Value:same directory as dbpath

Directory where Percona TokuMX will place temporary files used by the Bulk Loader for building indexes.

variable txnMemLimit
Default Value:1048576 bytes (1MB)
Values:up to 2097152 (2MB)

The amount, in bytes, of replication data a transaction stores in memory before spilling the data to disk.

This parameter only affects replica sets.

We recommend leaving the default unless there is some compelling reason to change this.

3.2. Advanced Parameters

Several server parameters can be set with --setParameter on the command line.

variable compressBuffersBeforeEviction
Default Value:true

Supported since 1.5.0

During partial eviction, we have the option to compress nonleaf node buffers rather than completely evicting them the first time we run partial eviction. This costs some CPU and doesn’t reclaim as much space, but may avoid disk I/O later if the buffer is reused soon.

variable defaultCompression

Supported since 1.4.2

Sets the default compression for new collections and indexes.

variable defaultFanout

Supported since 1.4.2

Sets the default fanout for new collections and indexes.

variable defaultPageSize

Supported since 1.4.2

Sets the default pageSize for new collections and indexes.

variable defaultReadPageSize

Supported since 1.4.2

Sets the default readPageSize for new collections and indexes.

variable electionBackoffMillis
Default Value:1000 ms

Supported since 2.0.0

During a replication election, multiple members may be electable. To reduce the chance of multiple elections running concurrently, basic MongoDB secondaries sleep for a random amount of time up to 1000 ms.

Percona TokuMX makes this duration configurable. Generally, this value should not be modified unless experiments show that concurrent elections often conflict and cause downtime. This can happen in replica sets with high-latency connections between members, especially in cross-datacenter scenarios. In that case, increasing electionBackoffMillis may reduce downtime due to concurrent elections.

variable forceWriteLocks

:default:false

Supported since 1.5.0

If this setting is true, then when a client is waiting to acquire one of the per-database Metadata Locks, all other clients that hold the read lock will abort and need to be retried. This can avoid stalls in some workloads, but requires that the application understand it may need to retry some operations.

This setting controls the default value for setWriteLockYielding for each newly-created connection.

variable loaderCompressTmp
Default Value:true

Supported since 1.4.0

Controls whether the Bulk Loader compresses temporary files put in tmpDir with quicklz.

variable migrateStartCloneLockTimeout
Default Value:60000 ms

Supported since 1.5.0

Controls how long a sharding chunk migration will wait to acquire a range lock on the chunk it intends to move.

variable migrateUniqueChecks

:default:true

Supported since 1.4.0

Controls whether sharding chunk migrations do unique checks when inserting new data on the recipient shard.

This setting is true by default, which is the safest option, but if you know that any of the following are true, you can turn this off for a performance boost during chunk migrations:

  • You are sharding with a primaryKey.
  • You are allowing MongoDB to assign _ids for your application.
  • You know your _ids are all globally unique and are willing to accept the consequences if they aren’t.

For a full discussion of this setting’s meaning, see What’s new in TokuMX 1.4, Part 5: Faster chunk migrations.

variable soTimeoutForReplLargeTxn
Default Value:600s

Supported since 1.5.1

Specifies the socket timeout of replication thread’s connection while replicating a large transaction stored in local.oplog.refs.

variable numCachetableBucketMutexes
Default Value:1000000

Supported since 1.3.2

Internal use only.

variable pkUniqueChecks
Default Value:true

Supported since 1.5.0

Internal use only.

3.4. Deprecated Parameters

Some parameters available in basic MongoDB have been deprecated in Percona TokuMX.

  • directoryperdb

All data is in one directory.

  • dur/journal

Crash safety is always on.

  • durOptions/journalOptions

Currently, there are no options with respect to crash safety.

  • nodur/nojournal

Crash safety cannot be disabled.

  • smallfiles

Percona TokuMX does not preallocate data files nearly as aggressively as MongoDB.

  • noprealloc/nopreallocj

We do not aggressively preallocate indexes or logs.

  • repair

We automatically recover indexes to a consistent state.

  • upgrade

We automatically upgrade indexes to the current format.

  • nssize

Our implementation of the namespace index is not fixed-size.

  • oplogSize

The oplog is no longer a capped collection.

  • master

Classic master/slave replication from MongoDB 1.x is not supported.

  • slave

Classic master/slave replication from MongoDB 1.x is not supported.

  • pretouch

Classic master/slave replication from MongoDB 1.x is not supported.

  • replIndexPrefetch

Replication does not require prefetching, and therefore no longer has prefetch threads.

Download PDF Manual

Table Of Contents

This Page