ScyllaDB Docs ScyllaDB Open Source CQL Reference Compaction

Caution

You're viewing documentation for a previous version. Switch to the latest stable version.

Compaction¶

This document describes the compaction strategy options available when creating a table. For more information about creating a table in Scylla, refer to the CQL Reference.

By default, Scylla starts a compaction task whenever a new SSTable is written. Compaction merges several SSTables into a new SSTable, which contains only the live data from the input SSTables. Merging several sorted files to get a sorted result is an efficient process, and this is the main reason why SSTables are kept sorted.

The following compaction strategies are supported by Scylla:

Size-tiered Compaction Strategy (STCS)
Leveled Compaction Strategy (LCS)
Incremental Compaction Strategy (ICS)
Time-window Compaction Strategy (TWCS)
Date-tiered Compaction Strategy (DTCS) - use TWCS instead

This page concentrates on the parameters to use when creating a table with a compaction strategy. If you are unsure which strategy to use or want general information on the compaction strategies which are available to Scylla, refer to Compaction Strategies.

Common options¶

The following options are available for all compaction strategies.

compaction = {
  'class' : 'compaction_strategy_name',
  'enabled' : (true | false),
  'tombstone_threshold' : ratio,
  'tombstone_compaction_interval' : sec}

class (default: SizeTieredCompactionStrategy)

Selects the compaction strategy.

It can be one of the following. If you are unsure which one to choose, refer to Compaction Strategies :

SizeTieredCompactionStrategy
TimeWindowCompactionStrategy
LeveledCompactionStrategy

enabled (default: true)

Runs background compaction (known as “minor compaction”). Can be one of the following:

true - runs minor compaction
false - disable minor compaction

tombstone_threshold (default: 0.2): The ratio (expressed as a decimal) of garbage-collectable tombstones compared to the data. When this threshold is exceeded on a specific table, a single SSTable compaction begins. Acceptable values are numbers in the range 0 -1.

tombstone_compaction_interval (default: 86400s (1 day)): An SSTable that is suitable for single SSTable compaction, according to tombstone_threshold will not be compacted if it is newer than tombstone_compaction_interval.

Size Tiered Compaction Strategy (STCS)¶

When using STCS, SSTables are put in different buckets depending on their size. When an SSTable is bucketed, the average size of the tables is compared to the new table as well as the high and low threshold levels.

The database compares each SSTable size to the average of all SSTable sizes on the node. It calculates bucket_low * avg_bucket_size and bucket_high * avg_bucket_size and then compares the result with the average SSTable size. The conditions set for bucket_high and bucket_low dictate if successive tables will be added to the same bucket. When compaction begins it merges SSTables whose size in KB are within [average-size * bucket_low] and [average-size * bucket_high].

Once the min_threashold is reached, minor compaction begins.

STCS options¶

The following options only apply to SizeTieredCompactionStrategy:

compaction = {
  'class' : 'SizeTieredCompactionStrategy',
  'bucket_high' : factor,
  'bucket_low' : factor,
  'min_sstable_size' : int,
  'min_threshold' : num_sstables,
  'max_threshold' : num_sstables}

bucket_high (default: 1.5)

A new SSTable is added to the bucket if the SSTable size is less than bucket_high * the average size of that bucket (and if the bucket_low condition also holds).

For example, if ‘bucket_high = 1.5’ and the SSTable size = 14MB, does it belong to a bucket with an average size of 10MB?

Yes, because the SSTable size = 14`, which is less than ‘bucket_high’ * average bucket size = 15.

So, the SSTable will be added to the bucket, and the bucket’s average size will be recalculated.

bucket_low (default: 0.5)

A new SSTable is added to the bucket if the SSTable size is more than bucket_low* the average size of that bucket (and if the bucket_high condition also holds).

For example, if ‘bucket_high = 0.5’ and the SSTable size is 10MB, does it belong to a bucket with an average size is 15MB?

Yes, because the SSTable size = 10 which is more than ‘bucket_low’ * average bucket size = 7.5.

So, the SSTable will be added to the bucket, and the bucket’s average size will be recalculated.

min_sstable_size (default: 50): All SSTables smaller than this number of bytes are put into the same bucket.

min_threshold (default: 4): Minimum number of SSTables that need to belong to the same size bucket before compaction is triggered on that bucket. If your SSTables are small, use min_sstable_size to define a size threshold (in bytes) below which all SSTables belong to one unique bucket.

Note

Enforcement of min_threshold is controlled by the compaction_enforce_min_threshold configuration option in the scylla.yaml configuration settings. By default, compaction_enforce_min_threshold: false, meaning the Size-Tiered Compaction Strategy will compact any bucket containing 2 or more SSTables. Otherwise, if compaction_enforce_min_threshold: true, the value of min_threshold is considered and only those buckets that contain at least min_threshold SSTables will be compacted.

max_threshold (default: 32): Maximum number of SSTables that will be compacted together in one compaction step.

Leveled Compaction Strategy (LCS)¶

The compaction class LeveledCompactionStrategy (LCS) creates SSTables of a fixed, relatively small size (160 MB by default) that are grouped into levels. Within each level, SSTables are guaranteed to be non-overlapping. Each level (L0, L1, L2 and so on) is ten times as large as the previous level.

LCS options¶

compaction = {
  'class' : 'LeveledCompactionStrategy',
  'sstable_size_in_mb' : int}

sstable_size_in_mb (default: 160): This is the target size in megabytes, that will be used as the goal for an SSTable size following a compression. Although SSTable sizes should be less or equal to sstable_size_in_mb, it is possible that compaction could produce a larger SSTable during compaction. This occurs when data for a given partition key is exceptionally large.

Incremental Compaction Strategy (ICS)¶

Added in version 2019.1.4: Scylla Enterprise

Note

ICS is only available for Scylla Enterprise customers

When using ICS, SSTable runs are put in different buckets depending on their size. When an SSTable run is bucketed, the average size of the runs in the bucket is compared to the new run, as well as the bucket_high and bucket_low levels.

The database compares each SSTable-run size to the average of all SSTable-run sizes on all buckets in the node. It calculates bucket_low * avg_bucket_size and bucket_high * avg_bucket_size and then compares the result with the average SSTable-run size. The conditions set for bucket_high and bucket_low dictate if successive runs will be added to the same bucket. When compaction begins it merges SSTable runs whose size in KB are within [average-size * bucket_low] and [average-size * bucket_high].

Once there are multiple runs in a bucket, minor compaction begins. The minimum number of SSTable runs that triggers minor compaction is either 2 or min_threshold, if the compaction_enforce_min_threshold configuration option is set in the scylla.yaml configuration file.

ICS options¶

The following options only apply to IncrementalCompactionStrategy:

compaction = {
  'class' : 'IncrementalCompactionStrategy',
  'bucket_high' : factor,
  'bucket_low' : factor,
  'min_sstable_size' : int,
  'min_threshold' : num_sstables,
  'max_threshold' : num_sstables,
  'sstable_size_in_mb' : int,
  'space_amplification_goal' : double}

bucket_high (default: 1.5): A new SSTable is added to the bucket if the SSTable size is less than bucket_high * the average size of that bucket (and if the bucket_low condition also holds). For example, if ‘bucket_high = 1.5’ and the SSTable size = 14MB, does the SSTable belong to a bucket with an average size of 10MB? Yes, because the SSTable size = 14, which is less than ‘bucket_high’ * average bucket size = 15. So, the SSTable will be added to the bucket, and the bucket’s average size will be recalculated.

bucket_low (default: 0.5): A new SSTable is added to the bucket if the SSTable size is more than bucket_low * the average size of that bucket (and if the bucket_high condition also holds). For example, if ‘bucket_high = 0.5’ and the SSTable size is 10MB, does the SSTable belong to a bucket with an average size of 15MB? Yes, because the SSTable size = 10, which is more than ‘bucket_low’ * average bucket size = 7.5. So, the SSTable will be added to the bucket, and the bucket’s average size will be recalculated.

min_sstable_size (default: 50)

All SSTables smaller than this number of megabytes are put into the same bucket.

Unlike Apache Cassandra, scylla uses uncompressed size when bucketing similar-sized tiers together. Since compaction works on uncompressed data, SSTables containing similar amounts of data should be compacted together, even when they have different compression ratios.

min_threshold (default: 4): Minimum number of SSTable runs that need to belong to the same size bucket before compaction is triggered on that bucket.

Note

Enforcement of min_threshold is controlled by the compaction_enforce_min_threshold configuration option in the scylla.yaml configuration settings. By default, compaction_enforce_min_threshold=false, meaning the Incremental Compaction Strategy will compact any bucket containing 2 or more SSTable runs. Otherwise, if compaction_enforce_min_threshold=true, the value of min_threshold is considered and only those buckets that contain at least min_threshold SSTable runs will be compacted.

max_threshold (default: 32): Maximum number of SSTables that will be compacted together in one compaction step.

sstable_size_in_mb (default: 1000): This is the target size in megabytes, that will be used as the goal for an SSTable size (fragment size) following a compression.

space_amplification_goal (default: null)

ScyllaDB Enterprise

Added in version 2020.1.6.

This is a threshold of the ratio of the sum of the sizes of the two largest tiers to the size of the largest tier, above which ICS will automatically compact the second largest and largest tiers together to eliminate stale data that may have been overwritten, expired, or deleted. The space_amplification_goal is given as a double-precision floating point number that must be greater than 1.0.

For example, if ‘space_amplification_goal = 1.25’ and the largest tier holds 1000GB, when the second-largest tier accumulates SSTables with the total size of 250GB or more, the space_amplification_goal threshold is crossed and all the SSTables in the largest and second-largest tiers will be compacted together.

Time Window CompactionStrategy (TWCS)¶

The basic concept is that TimeWindowCompactionStrategy will create 1 SSTable per file for a given time window.

Caution

We strongly recommend using a single TTL value for any given table.
This means sticking to the default time to live as specified in the table’s schema.
Using multiple TTL values for a given table may lead to inefficiency when purging expired data, because an SSTable will remain until all of its data is expired.
Tombstone compaction can be enabled to remove data from partially expired SSTables, but this creates additional WA (write amplification).

Caution

Avoid overwriting data and deleting data explicitly at all costs, as this can potentially block an expired SSTable from being purged, due to the checks that are performed to avoid data resurrection.

TWCS options¶

compaction = {
  'class' : 'TimeWindowCompactionStrategy',
  'compaction_window_unit' : string,
  'compaction_window_size' : int,
  'expired_sstable_check_frequency_seconds' : int,
  'min_threshold' : num_sstables,
  'max_threshold' : num_sstables}

compaction_window_unit (default: DAYS)

A time unit used to determine the window size which can be one of the following:

'MINUTES'
'HOURS'
'DAYS'

compaction_window_size (default: 1): The number of units which will make up a window.

expired_sstable_check_frequency_seconds (default: 600): Specifies (in seconds) how often Scylla will check for fully expired SSTables, which can be immediately dropped.

min_threshold (default: 4): Minimum number of SSTables that need to belong to the same size bucket before compaction is triggered on that bucket.

max_threshold (default: 32): Maximum number of SSTables that will be compacted together in one compaction step.

Compaction¶

Common options¶

Size Tiered Compaction Strategy (STCS)¶

STCS options¶

Leveled Compaction Strategy (LCS)¶

LCS options¶

Incremental Compaction Strategy (ICS)¶

ICS options¶

Time Window CompactionStrategy (TWCS)¶

TWCS options¶

See Also¶