All accumulo properties have a default value in the source code. Properties can also be set in accumulo-site.xml and in zookeeper on per-table or system-wide basis. If properties are set in more than one location, accumulo will choose the property with the highest precedence. This order of precedence is described below (from highest to lowest):
Location | Description |
---|---|
Zookeeper table properties |
Table properties are applied to the entire cluster when set in zookeeper using the accumulo API or shell. While table properties take precedent over system properties, both will override properties set in accumulo-site.xml Table properties consist of all properties with the table.* prefix. Table properties are configured on a per-table basis using the following shell commmand: config -t TABLE -s PROPERTY=VALUE |
Zookeeper system properties |
System properties are applied to the entire cluster when set in zookeeper using the accumulo API or shell. System properties consist of all properties with a 'yes' in the 'Zookeeper Mutable' column in the table below. They are set with the following shell command:
config -s PROPERTY=VALUEIf a table.* property is set using this method, the value will apply to all tables except those configured on per-table basis (which have higher precedence). While most system properties take effect immediately, some require a restart of the process which is indicated in 'Zookeeper Mutable'. |
accumulo-site.xml | Accumulo processes (master, tserver, etc) read their local accumulo-site.xml on start up. Therefore, changes made to accumulo-site.xml must rsynced across the cluster and processes must be restarted to apply changes. Certain properties (indicated by a 'no' in 'Zookeeper Mutable') cannot be set in zookeeper and only set in this file. The accumulo-site.xml also allows you to configure tablet servers with different settings. |
Default | All properties have a default value in the source code. This value has the lowest precedence and is overriden if set in accumulo-site.xml or zookeeper. While the default value is usually optimal, there are cases where a change can increase query and ingest performance. |
The 'config' command in the shell allows you to view the current system configuration. You can also use the '-t' option to view a table's configuration as below:
$ ./bin/accumulo shell -u root Enter current password for 'root'@'ac13': ****** Shell - Apache Accumulo Interactive Shell - - version: 1.3.6 - instance name: ac13 - instance id: 4f48fa03-f692-43ce-ae03-94c9ea8b7181 - - type 'help' for a list of available commands - root@ac13> config -t foo ---------+---------------------------------------------+------------------------------------------------------ SCOPE | NAME | VALUE ---------+---------------------------------------------+------------------------------------------------------ default | table.balancer ............................ | org.apache.accumulo.server.master.balancer.DefaultLoadBalancer default | table.bloom.enabled ....................... | false default | table.bloom.error.rate .................... | 0.5% default | table.bloom.hash.type ..................... | murmur default | table.bloom.key.functor ................... | org.apache.accumulo.core.file.keyfunctor.RowFunctor default | table.bloom.load.threshold ................ | 1 default | table.bloom.size .......................... | 1048576 default | table.cache.block.enable .................. | false default | table.cache.index.enable .................. | false default | table.compaction.major.everything.at ...... | 19700101000000GMT default | table.compaction.major.everything.idle .... | 1h default | table.compaction.major.ratio .............. | 1.3 site | @override .............................. | 1.4 system | @override .............................. | 1.5 table | @override .............................. | 1.6 default | table.compaction.minor.idle ............... | 5m default | table.compaction.minor.logs.threshold ..... | 3 default | table.failures.ignore ..................... | false
Jump to: instance.* | general.* | master.* | tserver.* | logger.* | gc.* | monitor.* | trace.* | table.* | table.constraint.* | table.iterator.* | table.group.*
instance.* | ||||
Properties in this category must be consistent throughout a cloud. This is enforced and servers won't be able to communicate if these differ. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
---|---|---|---|---|
instance.dfs.dir | absolute path | no | /accumulo |
HDFS directory in which accumulo instance will run. Do not change after accumulo is initialized. |
instance.secret | string | no | DEFAULT |
A secret unique to a given instance that all servers must know in order to communicate with one another. Do not change after accumulo is initialized. |
instance.zookeeper.host | host list | no | localhost:2181 |
Comma separated list of zookeeper servers |
instance.zookeeper.timeout | duration | no | 30s |
Zookeeper session timeout; max value when represented as milliseconds should be no larger than 2147483647 |
general.* | ||||
Properties in this category affect the behavior of accumulo overall, but do not have to be consistent throughout a cloud. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
general.classpaths | string | no | $ACCUMULO_HOME/src/server/target/classes/, $ACCUMULO_HOME/src/core/target/classes/, $ACCUMULO_HOME/src/start/target/classes/, $ACCUMULO_HOME/src/examples/target/classes/, $ACCUMULO_HOME/conf, $ACCUMULO_HOME/lib/[^.].$ACCUMULO_VERSION.jar, $ACCUMULO_HOME/lib/[^.].*.jar, $ZOOKEEPER_HOME/zookeeper[^.].*.jar, $HADOOP_HOME/[^.].*.jar, $HADOOP_HOME/conf, $HADOOP_HOME/lib/[^.].*.jar, |
A list of all of the places to look for a class. Order does matter, as it will look for the jar starting in the first location to the last. Please note, hadoop conf and hadoop lib directories NEED to be here, along with accumulo lib and zookeeper directory. Supports full regex on filename alone. |
general.dynamic.classpaths | string | no | $ACCUMULO_HOME/lib/ext/[^.].*.jar |
A list of all of the places where changes in jars or classes will force a reload of the classloader. |
general.rpc.timeout | duration | no | 120s |
Time to wait on I/O for simple, short RPC calls |
master.* | ||||
Properties in this category affect the behavior of the master server | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
master.logger.balancer | java class | yes | org.apache.accumulo.server.master.balancer.SimpleLoggerBalancer |
The balancer class that accumulo will use to make logger assignment decisions. |
master.port.client | port | yes but requires restart | 9999 |
The port used for handling client connections on the master |
master.recovery.max.age | duration | yes | 60m |
Recovery files older than this age will be removed. |
master.recovery.pool | string | yes | recovery |
Priority queue to use for log recovery map/reduce jobs. |
master.recovery.queue | string | yes | default |
Priority queue to use for log recovery map/reduce jobs. |
master.recovery.reducers | count | yes | 10 |
Number of reducers to use to sort recovery logs (per log) |
master.recovery.sort.mapreduce | boolean | yes | false |
If true, use map/reduce to sort write-ahead logs during recovery |
master.recovery.time.max | duration | yes | 30m |
The maximum time to attempt recovery before giving up |
master.tablet.balancer | java class | yes | org.apache.accumulo.server.master.balancer.DefaultLoadBalancer |
The balancer class that accumulo will use to make tablet assignment and migration decisions. |
tserver.* | ||||
Properties in this category affect the behavior of the tablet servers | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
tserver.bloom.load.concurrent.max | count | yes | 4 |
The number of concurrent threads that will load bloom filters in the background. Setting this to zero will make bloom filters load in the foreground. |
tserver.cache.data.size | memory | yes | 100M |
Specifies the size of the cache for file data blocks. |
tserver.cache.index.size | memory | yes | 50M |
Specifies the size of the cache for file indices. |
tserver.client.timeout | duration | yes | 3s |
Time to wait for clients to continue scans before closing a session. |
tserver.compaction.major.concurrent.max | count | yes but requires restart | 3 |
The maximum number of concurrent major compactions for a tablet server |
tserver.compaction.major.delay | duration | yes | 30s |
Time a tablet server will sleep between checking which tablets need compaction. |
tserver.compaction.major.files.open.max | count | yes but requires restart | 90 |
Max number of files a major compaction can open at once. At runtime this number is divided by the concurrent number of compactors. |
tserver.compaction.minor.concurrent.max | count | yes | 4 |
The maximum number of concurrent minor compactions for a tablet server |
tserver.default.blocksize | memory | yes | 1M |
Specifies a default blocksize for the tserver caches |
tserver.dir.memdump | path | yes | /tmp |
A long running scan could possibly hold memory that has been minor compacted. To prevent this, the in memory map is dumped to a local file and the scan is switched to that local file. We can not switch to the minor compacted file because it may have been modified by iterators. The file dumped to the local dir is an exact copy of what was in memory. |
tserver.files.open.idle | duration | yes | 1m |
Tablet servers leave previously used map files open for future queries. This setting determines how much time an unused map file should be kept open until it is closed. |
tserver.files.open.max | count | yes but requires restart | 150 |
Maximum total map files that all tablets in a tablet server can open. This includes major compactions. So the number of map files that can be opened for searches is: tserver.files.open.max - tserver.compaction.major.files.open.max |
tserver.logger.count | count | yes but requires restart | 2 |
The number of loggers that each tablet server should use. |
tserver.logger.strategy | string | yes | org.apache.accumulo.server.tabletserver.log.RoundRobinLoggerStrategy |
The classname used to decide which loggers to use. |
tserver.logger.timeout | duration | yes | 30s |
The time to wait for a logger to respond to a write-ahead request |
tserver.memory.lock | boolean | yes | false |
The tablet server must communicate with zookeeper frequently to maintain its locks. If the tablet server's memory is swapped out the java garbage collector can stop all processing for long periods. Change this property to true and the tablet server will attempt to lock all of its memory to RAM, which may reduce delays during java garbage collection. You will have to modify the system limit for "max locked memory". This feature is only available when running on Linux. Alternatively you may also want to set /proc/sys/vm/swappiness to zero (again, this is Linux-specific). |
tserver.memory.manager | java class | yes | org.apache.accumulo.server.tabletserver.LargestFirstMemoryManager |
An implementation of MemoryManger that accumulo will use. |
tserver.memory.maps.max | memory | yes | 1G |
Maximum amount of memory all tablets in memory maps can use. |
tserver.memory.maps.native.enabled | boolean | yes but requires restart | true |
An in-memory data store for accumulo implemented in c++ that increases the amount of data accumulo can hold in memory and avoids Java GC pauses. |
tserver.metadata.readahead.concurrent.max | count | yes | 8 |
The maximum number of concurrent metadata read ahead that will execute. |
tserver.migrations.concurrent.max | count | yes | 1 |
The maximum number of concurrent tablet migrations for a tablet server |
tserver.monitor.fs | boolean | yes | true |
When enabled the tserver will monitor file systems and kill itself when one switches from rw to ro. This is usually and indication that Linux has detected a bad disk. |
tserver.mutation.queue.max | memory | yes | 256K |
The amount of memory to use to store write-ahead-log mutations-per-session before flushing them. |
tserver.port.client | port | yes but requires restart | 9997 |
The port used for handling client connections on the tablet servers |
tserver.port.search | boolean | yes | false |
if the ports above are in use, search higher ports until one is available |
tserver.readahead.concurrent.max | count | yes | 16 |
The maximum number of concurrent read ahead that will execute. This effectivelylimits the number of long running scans that can run concurrently per tserver. |
tserver.session.idle.max | duration | yes | 1m |
maximum idle time for a session |
tserver.tablet.split.midpoint.files.max | count | yes | 30 |
To find a tablets split points, all index files are opened. This setting determines how many index files can be opened at once. When there are more index files than this setting multiple passes must be made, which is slower. However opening too many files at once can cause problems. |
tserver.walog.max.size | memory | yes | 1G |
The maximum size for each write-ahead log |
logger.* | ||||
Properties in this category affect the behavior of the write-ahead logger servers | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
logger.archive | boolean | yes | false |
determines if write-ahead logs are archived in hdfs |
logger.copy.threadpool.size | count | yes | 2 |
size of the thread pool used to copy files from the local log area to HDFS |
logger.dir.walog | path | yes | walogs |
The directory used to store write-ahead logs on the local filesystem |
logger.monitor.fs | boolean | yes | true |
When enabled the logger will monitor file systems and kill itself when one switches from rw to ro. This is usually and indication that Linux has detected a bad disk. |
logger.port.client | port | yes but requires restart | 11224 |
The port used for write-ahead logger services |
logger.port.search | boolean | yes | false |
if the port above is in use, search higher ports until one is available |
logger.sort.buffer.size | memory | yes | 200M |
The amount of memory to use when sorting logs during recovery. Only used when *not* sorting logs with map/reduce. |
gc.* | ||||
Properties in this category affect the behavior of the accumulo garbage collector. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
gc.cycle.delay | duration | yes | 5m |
Time between garbage collection cycles. In each cycle, old files no longer in use are removed from the filesystem. |
gc.cycle.start | duration | yes | 30s |
Time to wait before attempting to garbage collect any old files. |
gc.port.client | port | yes but requires restart | 50091 |
The listening port for the garbage collector's monitor service |
gc.threads.delete | count | yes | 16 |
The number of threads used to delete files |
monitor.* | ||||
Properties in this category affect the behavior of the monitor web server. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
monitor.port.client | port | no | 50095 |
The listening port for the monitor's http service |
monitor.port.log4j | port | no | 4560 |
The listening port for the monitor's log4j logging collection. |
trace.* | ||||
Properties in this category affect the behavior of distributed tracing. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
trace.password | string | no | secret |
The password for the user used to store distributed traces |
trace.port.client | port | no | 12234 |
The listening port for the trace server |
trace.table | string | no | trace |
The name of the table to store distributed traces |
trace.user | string | no | root |
The name of the user to store distributed traces |
table.* | ||||
Properties in this category affect tablet server treatment of tablets, but can be configured on a per-table basis. Setting these properties in the site file will override the default globally for all tables and not any specific table. However, both the default and the global setting can be overridden per table using the table operations API or in the shell, which sets the overridden value in zookeeper. Restarting accumulo tablet servers after setting these properties in the site file will cause the global setting to take effect. However, you must use the API or the shell to change properties in zookeeper that are set on a table. | ||||
Property | Type | Zookeeper Mutable | Default Value | Description |
table.balancer | string | yes | org.apache.accumulo.server.master.balancer.DefaultLoadBalancer |
This property can be set to allow the LoadBalanceByTable load balancer to change the called Load Balancer for this table |
table.bloom.enabled | boolean | yes | false |
Use bloom filters on this table. |
table.bloom.error.rate | fraction/percentage | yes | 0.5% |
Bloom filter error rate. |
table.bloom.hash.type | string | yes | murmur |
The bloom filter hash type |
table.bloom.key.functor | java class | yes | org.apache.accumulo.core.file.keyfunctor.RowFunctor |
A function that can transform the key prior to insertion and check of bloom filter. org.apache.accumulo.core.file.keyfunctor.RowFunctor,,org.apache.accumulo.core.file.keyfunctor.ColumnFamilyFunctor, and org.apache.accumulo.core.file.keyfunctor.ColumnQualifierFunctor are allowable values. One can extend any of the above mentioned classes to perform specialized parsing of the key. |
table.bloom.load.threshold | count | yes | 1 |
This number of seeks that would actually use a bloom filter must occur before a map files bloom filter is loaded. Set this to zero to initiate loading of bloom filters when a map file opened. |
table.bloom.size | count | yes | 1048576 |
Bloom filter size, as number of keys. |
table.cache.block.enable | boolean | yes | false |
Determines whether file block cache is enabled. |
table.cache.index.enable | boolean | yes | false |
Determines whether index cache is enabled. |
table.compaction.major.everything.at | date/time | yes | 19700101000000GMT |
This setting specifies a time at which all tablets in a table will major compact to one file, even tablets with only one file. When this settings specifies a time in the future, no action is taken. When the time is in the past any tablet having a map file older than the specified time will major compact to one file. The time specified must conform to the yyyyMMddHHmmssz pattern. See the Java SimpleDataFormat java doc for details about this pattern. |
table.compaction.major.everything.idle | duration | yes | 1h |
After a tablet has been idle (no mutations) for this time period it may have all of its map file compacted into one. There is no guarantee an idle tablet will be compacted. Compactions of idle tablets are only started when regular compactions are not running. Idle compactions only take place for tablets that have one or more map files. |
table.compaction.major.ratio | fraction/percentage | yes | 3 |
minimum ratio of total input size to maximum input file size for running a major compaction |
table.compaction.minor.idle | duration | yes | 5m |
After a tablet has been idle (no mutations) for this time period it may have its in-memory map flushed to disk in a minor compaction. There is no guarantee an idle tablet will be compacted. |
table.compaction.minor.logs.threshold | count | yes | 3 |
When there are more than this many write-ahead logs against a tablet, it will be minor compacted. |
table.failures.ignore | boolean | yes | false |
If you want queries for your table to hang or fail when data is missing from the system, then set this to false. When this set to true missing data will be reported but queries will still run possibly returning a subset of the data. |
table.file.blocksize | memory | yes | 0B |
Overrides the hadoop dfs.block.size setting so that map files have better query performance. The maximum value for this is 2147483647 |
table.file.compress.blocksize | memory | yes | 100K |
Overrides the hadoop io.seqfile.compress.blocksize setting so that map files have better query performance. The maximum value for this is 2147483647 |
table.file.compress.type | string | yes | gz |
One of gz,lzo,none |
table.file.replication | count | yes | 0 |
Determines how many replicas to keep of a tables map files in HDFS. When this value is LTE 0, HDFS defaults are used. |
table.file.type | string | yes | rf |
Change the type of file a table writes |
table.groups.enabled | string | yes | A comma separated list of locality group names to enable for this table. | |
table.scan.cache.enable | boolean | yes | false |
Determines whether scan cache is enabled. |
table.scan.cache.size | memory | yes | 8M |
Scan cache size. |
table.scan.max.memory | memory | yes | 1M |
The maximum amount of memory that will be used to cache results of a client query/scan. Once this limit is reached, the buffered data is sent to the client. |
table.security.scan.visibility.default | string | yes | The security label that will be assumed at scan time if an entry does not have a visibility set. Note: An empty security label is displayed as []. The scan results will show an empty visibility even if the visibility from this setting is applied to the entry. CAUTION: If a particular key has an empty security label AND its table's default visibility is also empty, access will ALWAYS be granted for users with permission to that table. Additionally, if this field is changed, all existing data with an empty visibility label will be interpreted with the new label on the next scan. |
|
table.split.threshold | memory | yes | 1G |
When combined size of mapfiles exceeds this amount a tablet is split. |
table.walog.enabled | boolean | yes | true |
Use the write-ahead log to prevent the loss of data. |
table.constraint.* | ||||
Properties in this category are per-table properties that add constraints to a table. These properties start with the category prefix, followed by a number, and their values correspond to a fully qualified Java class that implements the Constraint interface. For example, table.constraint.1 = org.apache.accumulo.core.constraints.MyCustomConstraint and table.constraint.2 = my.package.constraints.MySecondConstraint | ||||
table.iterator.* | ||||
Properties in this category specify iterators that are applied at various stages (scopes) of interaction with a table. These properties start with the category prefix, followed by a scope (minc, majc, scan, etc.), followed by a period, followed by a name, as in table.iterator.scan.vers, or table.iterator.scan.custom. The values for these properties are a number indicating the ordering in which it is applied, and a class name such as table.iterator.scan.vers = 10,org.apache.accumulo.core.iterators.VersioningIterator These iterators can take options if additional properties are set that look like this property, but are suffixed with a period, followed by 'opt' followed by another period, and a property name. For example, table.iterator.minc.vers.opt.maxVersions = 3 | ||||
table.group.* | ||||
Properties in this category are per-table properties that define locality groups in a table. These properties start with the category prefix, followed by a name, followed by a period, and followed by a property for that group. For example table.group.group1=x,y,z sets the column families for a group called group1. Once configured, group1 can be enabled by adding it to the list of groups in the table.groups.enabled property. Additional group options may be specified for a named group by setting table.group.<name>.opt.<key>=<value>. |
Property Type | Description |
---|---|
duration |
A non-negative integer optionally followed by a unit of time (whitespace disallowed), as in 30s. If no unit of time is specified, seconds are assumed. Valid units are 'ms', 's', 'm', 'h' for milliseconds, seconds, minutes, and hours. Examples of valid durations are '600', '30s', '45m', '30000ms', '3d', and '1h'. Examples of invalid durations are '1w', '1h30m', '1s 200ms', 'ms', '', and 'a'. Unless otherwise stated, the max value for the duration represented in milliseconds is 9223372036854775807 |
date/time |
A date/time string in the format: YYYYMMDDhhmmssTTT where TTT is the 3 character time zone |
memory |
A positive integer optionally followed by a unit of memory (whitespace disallowed), as in 2G. If no unit is specified, bytes are assumed. Valid units are 'B', 'K', 'M', 'G', for bytes, kilobytes, megabytes, and gigabytes. Examples of valid memories are '1024', '20B', '100K', '1500M', '2G'. Examples of invalid memories are '1M500K', '1M 2K', '1MB', '1.5G', '1,024K', '', and 'a'. Unless otherwise stated, the max value for the memory represented in bytes is 9223372036854775807 |
host list |
A comma-separated list of hostnames or ip addresses, with optional port numbers. Examples of valid host lists are 'localhost:2000,www.example.com,10.10.1.1:500' and 'localhost'. Examples of invalid host lists are '', ':1000', and 'localhost:80000' |
port |
An positive integer in the range 1024-65535, not already in use or specified elsewhere in the configuration |
count |
A non-negative integer in the range of 0-2147483647 |
fraction/percentage |
A floating point number that represents either a fraction or, if suffixed with the '%' character, a percentage. Examples of valid fractions/percentages are '10', '1000%', '0.05', '5%', '0.2%', '0.0005'. Examples of invalid fractions/percentages are '', '10 percent', 'Hulk Hogan' |
path |
A string that represents a filesystem path, which can be either relative or absolute to some directory. The filesystem depends on the property. |
absolute path |
An absolute filesystem path. The filesystem depends on the property. This is the same as path, but enforces that its root is explicitly specified. |
java class |
A fully qualified java class name representing a class on the classpath. An example is 'java.lang.String', rather than 'String' |
string |
An arbitrary string of characters whose format is unspecified and interpreted based on the context of the property to which it applies. |
boolean |
Has a value of either 'true' or 'false' |