writer-jill commented on code in PR #12344: URL: https://github.com/apache/druid/pull/12344#discussion_r891055918
########## docs/design/segments.md: ########## @@ -23,231 +23,198 @@ title: "Segments" --> -Apache Druid stores its index in *segment files*, which are partitioned by -time. In a basic setup, one segment file is created for each time +Apache Druid stores its index in *segment files* partitioned by +time. In a basic setup, Druid creates one segment file for each time interval, where the time interval is configurable in the `segmentGranularity` parameter of the -[`granularitySpec`](../ingestion/ingestion-spec.md#granularityspec). For Druid to -operate well under heavy query load, it is important for the segment +[`granularitySpec`](../ingestion/ingestion-spec.md#granularityspec). + +For Druid to operate well under heavy query load, it is important for the segment file size to be within the recommended range of 300MB-700MB. If your segment files are larger than this range, then consider either changing the granularity of the time interval or partitioning your -data and tweaking the `targetRowsPerSegment` in your `partitionsSpec` -(a good starting point for this parameter is 5 million rows). See the -sharding section below and the 'Partitioning specification' section of +data and adjusting the `targetRowsPerSegment` in your `partitionsSpec`. +A good starting point for this parameter is 5 million rows. + +See the Sharding section below and the 'Partitioning specification' section of the [Batch ingestion](../ingestion/hadoop.md#partitionsspec) documentation -for more information. +for more guidance. -### A segment file's core data structures +## Segment file structure -Here we describe the internal structure of segment files, which is -essentially *columnar*: the data for each column is laid out in -separate data structures. By storing each column separately, Druid can -decrease query latency by scanning only those columns actually needed -for a query. There are three basic column types: the timestamp -column, dimension columns, and metric columns, as illustrated in the -image below: +Segment files are *columnar*: the data for each column is laid out in +separate data structures. By storing each column separately, Druid decreases query latency by scanning only those columns actually needed for a query. There are three basic column types: timestamp, dimensions, and metrics:  -The timestamp and metric columns are simple: behind the scenes each of -these is an array of integer or floating point values compressed with -LZ4. Once a query knows which rows it needs to select, it simply -decompresses these, pulls out the relevant rows, and applies the -desired aggregation operator. As with all columns, if a query doesn’t -require a column, then that column’s data is just skipped over. +Timestamp and metrics type columns are arrays of integer or floating point values compressed with +[LZ4](https://github.com/lz4/lz4-java). Once a query identifies which rows to select, it decompresses them, pulls out the relevant rows, and applies the +desired aggregation operator. If a query doesn’t require a column, Druid skips over that column's data. -Dimensions columns are different because they support filter and +Dimension columns are different because they support filter and group-by operations, so each dimension requires the following three data structures: -1. A dictionary that maps values (which are always treated as strings) to integer IDs, -2. A list of the column’s values, encoded using the dictionary in 1, and -3. For each distinct value in the column, a bitmap that indicates which rows contain that value. - - -Why these three data structures? The dictionary simply maps string -values to integer ids so that the values in \(2\) and \(3\) can be -represented compactly. The bitmaps in \(3\) -- also known as *inverted -indexes* allow for quick filtering operations (specifically, bitmaps -are convenient for quickly applying AND and OR operators). Finally, -the list of values in \(2\) is needed for *group by* and *TopN* -queries. In other words, queries that solely aggregate metrics based -on filters do not need to touch the list of dimension values stored in \(2\). +- Dictionary: Maps values (which are always treated as strings) to integer IDs, allowing compact representation of the list and bitmap values. +- List: The column’s values, encoded using the dictionary. Required for GroupBy and TopN queries. These operators allow queries that solely aggregate metrics based on filters to run without accessing the list of values. +- Bitmap: One bitmap for each distinct value in the column, to indicate which rows contain that value. Bitmaps allow for quick filtering operations because they are convenient for quickly applying AND and OR operators. Also known as inverted indexes. -To get a concrete sense of these data structures, consider the ‘page’ -column from the example data above. The three data structures that -represent this dimension are illustrated in the diagram below. +To get a better sense of these data structures, consider the ‘page’ column from the given example data as represented by the following data structures: ``` -1: Dictionary that encodes column values - { +1: Dictionary + { "Justin Bieber": 0, "Ke$ha": 1 - } + } -2: Column data - [0, +2: List of column data + [0, 0, 1, 1] -3: Bitmaps - one for each unique value of the column - value="Justin Bieber": [1,1,0,0] - value="Ke$ha": [0,0,1,1] +3: Bitmaps + value="Justin Bieber": [1,1,0,0] + value="Ke$ha": [0,0,1,1] ``` -Note that the bitmap is different from the first two data structures: -whereas the first two grow linearly in the size of the data (in the -worst case), the size of the bitmap section is the product of data -size * column cardinality. Compression will help us here though -because we know that for each row in 'column data', there will only be a -single bitmap that has non-zero entry. This means that high cardinality -columns will have extremely sparse, and therefore highly compressible, -bitmaps. Druid exploits this using compression algorithms that are -specially suited for bitmaps, such as roaring bitmap compression. +Note that the bitmap is different from the dictionary and list data structures: the dictionary and list grow linearly with the size of the data, but the size of the bitmap section is the product of data size * column cardinality. Review Comment: Logged as issue: https://github.com/apache/druid/issues/12614 ########## docs/design/segments.md: ########## @@ -23,231 +23,198 @@ title: "Segments" --> -Apache Druid stores its index in *segment files*, which are partitioned by -time. In a basic setup, one segment file is created for each time +Apache Druid stores its index in *segment files* partitioned by +time. In a basic setup, Druid creates one segment file for each time interval, where the time interval is configurable in the `segmentGranularity` parameter of the -[`granularitySpec`](../ingestion/ingestion-spec.md#granularityspec). For Druid to -operate well under heavy query load, it is important for the segment +[`granularitySpec`](../ingestion/ingestion-spec.md#granularityspec). + +For Druid to operate well under heavy query load, it is important for the segment file size to be within the recommended range of 300MB-700MB. If your segment files are larger than this range, then consider either changing the granularity of the time interval or partitioning your -data and tweaking the `targetRowsPerSegment` in your `partitionsSpec` -(a good starting point for this parameter is 5 million rows). See the -sharding section below and the 'Partitioning specification' section of +data and adjusting the `targetRowsPerSegment` in your `partitionsSpec`. +A good starting point for this parameter is 5 million rows. + +See the Sharding section below and the 'Partitioning specification' section of the [Batch ingestion](../ingestion/hadoop.md#partitionsspec) documentation -for more information. +for more guidance. -### A segment file's core data structures +## Segment file structure -Here we describe the internal structure of segment files, which is -essentially *columnar*: the data for each column is laid out in -separate data structures. By storing each column separately, Druid can -decrease query latency by scanning only those columns actually needed -for a query. There are three basic column types: the timestamp -column, dimension columns, and metric columns, as illustrated in the -image below: +Segment files are *columnar*: the data for each column is laid out in +separate data structures. By storing each column separately, Druid decreases query latency by scanning only those columns actually needed for a query. There are three basic column types: timestamp, dimensions, and metrics:  -The timestamp and metric columns are simple: behind the scenes each of -these is an array of integer or floating point values compressed with -LZ4. Once a query knows which rows it needs to select, it simply -decompresses these, pulls out the relevant rows, and applies the -desired aggregation operator. As with all columns, if a query doesn’t -require a column, then that column’s data is just skipped over. +Timestamp and metrics type columns are arrays of integer or floating point values compressed with +[LZ4](https://github.com/lz4/lz4-java). Once a query identifies which rows to select, it decompresses them, pulls out the relevant rows, and applies the +desired aggregation operator. If a query doesn’t require a column, Druid skips over that column's data. -Dimensions columns are different because they support filter and +Dimension columns are different because they support filter and group-by operations, so each dimension requires the following three data structures: -1. A dictionary that maps values (which are always treated as strings) to integer IDs, -2. A list of the column’s values, encoded using the dictionary in 1, and -3. For each distinct value in the column, a bitmap that indicates which rows contain that value. - - -Why these three data structures? The dictionary simply maps string -values to integer ids so that the values in \(2\) and \(3\) can be -represented compactly. The bitmaps in \(3\) -- also known as *inverted -indexes* allow for quick filtering operations (specifically, bitmaps -are convenient for quickly applying AND and OR operators). Finally, -the list of values in \(2\) is needed for *group by* and *TopN* -queries. In other words, queries that solely aggregate metrics based -on filters do not need to touch the list of dimension values stored in \(2\). +- Dictionary: Maps values (which are always treated as strings) to integer IDs, allowing compact representation of the list and bitmap values. +- List: The column’s values, encoded using the dictionary. Required for GroupBy and TopN queries. These operators allow queries that solely aggregate metrics based on filters to run without accessing the list of values. +- Bitmap: One bitmap for each distinct value in the column, to indicate which rows contain that value. Bitmaps allow for quick filtering operations because they are convenient for quickly applying AND and OR operators. Also known as inverted indexes. -To get a concrete sense of these data structures, consider the ‘page’ -column from the example data above. The three data structures that -represent this dimension are illustrated in the diagram below. +To get a better sense of these data structures, consider the ‘page’ column from the given example data as represented by the following data structures: ``` -1: Dictionary that encodes column values - { +1: Dictionary + { "Justin Bieber": 0, "Ke$ha": 1 - } + } -2: Column data - [0, +2: List of column data + [0, 0, 1, 1] -3: Bitmaps - one for each unique value of the column - value="Justin Bieber": [1,1,0,0] - value="Ke$ha": [0,0,1,1] +3: Bitmaps + value="Justin Bieber": [1,1,0,0] + value="Ke$ha": [0,0,1,1] ``` -Note that the bitmap is different from the first two data structures: -whereas the first two grow linearly in the size of the data (in the -worst case), the size of the bitmap section is the product of data -size * column cardinality. Compression will help us here though -because we know that for each row in 'column data', there will only be a -single bitmap that has non-zero entry. This means that high cardinality -columns will have extremely sparse, and therefore highly compressible, -bitmaps. Druid exploits this using compression algorithms that are -specially suited for bitmaps, such as roaring bitmap compression. +Note that the bitmap is different from the dictionary and list data structures: the dictionary and list grow linearly with the size of the data, but the size of the bitmap section is the product of data size * column cardinality. + +For each row in the list of column data, there is only a single bitmap that has a non-zero entry. This means that high cardinality columns have extremely sparse, and therefore highly compressible, bitmaps. Druid exploits this using compression algorithms that are specially suited for bitmaps, such as [Roaring bitmap compression](https://github.com/RoaringBitmap/RoaringBitmap). + +## Handling null values + +By default, Druid string dimension columns use the values `''` and `null` interchangeably and numeric and metric columns can not represent `null` at all, instead coercing nulls to `0`. However, Druid also provides a SQL compatible null handling mode, which you can enable at the system level, through `druid.generic.useDefaultValueForNull`. This setting, when set to `false`, allows Druid to create segments _at ingestion time_ in which the string columns can distinguish `''` from `null`, and numeric columns which can represent `null` valued rows instead of `0`. + +String dimension columns contain no additional column structures in this mode, instead they reserve an additional dictionary entry for the `null` value. Numeric columns are stored in the segment with an additional bitmap in which the set bits indicate `null` valued rows. + +In addition to slightly increased segment sizes, SQL compatible null handling can incur a performance cost at query time, due to the need to check the null bitmap. This performance cost only occurs for columns that actually contain null values. + +## Segments with different schemas + +Druid segments for the same datasource may have different schemas. If a string column (dimension) exists in one segment but not another, queries that involve both segments still work. Queries for the segment without the dimension behave as if the dimension contains only null values. Similarly, if one segment has a numeric column (metric) but another does not, queries on the segment without the metric generally operate as expected. Aggregations over the missing metric operate as if the metric doesn't exist. + +## Column format + +Each column is stored as two parts: + +- A Jackson-serialized ColumnDescriptor. +- The rest of the binary for the column. + +A ColumnDescriptor is an object that allows the use of Jackson's polymorphic deserialization to add new and interesting methods of serialization with minimal impact to the code. It consists of some metadata about the column (for example: type, whether it's multi-value) and a list of serialization/deserialization logic that can deserialize the rest of the binary. ### Multi-value columns -If a data source makes use of multi-value columns, then the data -structures within the segment files look a bit different. Let's -imagine that in the example above, the second row were tagged with -both the 'Ke$ha' *and* 'Justin Bieber' topics. In this case, the three -data structures would now look as follows: +If a data source uses multi-value columns, then the data structures within the segment files look a bit different. Let's imagine that in the example above, the second row is tagged with both the `Ke$ha` *and* `Justin Bieber` topics, as follows: ``` -1: Dictionary that encodes column values - { +1: Dictionary + { "Justin Bieber": 0, "Ke$ha": 1 - } + } -2: Column data - [0, - [0,1], <--Row value of multi-value column can have array of values +2: List of column data + [0, + [0,1], <--Row value in a multi-value column can contain an array of values 1, 1] -3: Bitmaps - one for each unique value - value="Justin Bieber": [1,1,0,0] - value="Ke$ha": [0,1,1,1] +3: Bitmaps + value="Justin Bieber": [1,1,0,0] + value="Ke$ha": [0,1,1,1] ^ | | - Multi-value column has multiple non-zero entries + Multi-value column contains multiple non-zero entries ``` -Note the changes to the second row in the column data and the Ke$ha +Note the changes to the second row in the list of column data and the `Ke$ha` bitmap. If a row has more than one value for a column, its entry in -the 'column data' is an array of values. Additionally, a row with *n* -values in 'column data' will have *n* non-zero valued entries in -bitmaps. - -## SQL Compatible Null Handling -By default, Druid string dimension columns use the values `''` and `null` interchangeably and numeric and metric columns can not represent `null` at all, instead coercing nulls to `0`. However, Druid also provides a SQL compatible null handling mode, which must be enabled at the system level, through `druid.generic.useDefaultValueForNull`. This setting, when set to `false`, will allow Druid to _at ingestion time_ create segments whose string columns can distinguish `''` from `null`, and numeric columns which can represent `null` valued rows instead of `0`. - -String dimension columns contain no additional column structures in this mode, instead just reserving an additional dictionary entry for the `null` value. Numeric columns however will be stored in the segment with an additional bitmap whose set bits indicate `null` valued rows. In addition to slightly increased segment sizes, SQL compatible null handling can incur a performance cost at query time as well, due to the need to check the null bitmap. This performance cost only occurs for columns that actually contain nulls. - -## Naming Convention - -Identifiers for segments are typically constructed using the segment datasource, interval start time (in ISO 8601 format), interval end time (in ISO 8601 format), and a version. If data is additionally sharded beyond a time range, the segment identifier will also contain a partition number. - -An example segment identifier may be: -datasource_intervalStart_intervalEnd_version_partitionNum - -## Segment Components +the list is an array of values. Additionally, a row with *n* values in the list has *n* non-zero valued entries in bitmaps. -Behind the scenes, a segment is comprised of several files, listed below. +## Compression -* `version.bin` +Druid uses LZ4 by default to compress blocks of values for string, long, float, and double columns. Druid uses Roaring to compress bitmaps for string columns and numeric null values. We recommend that you use these defaults unless you've experimented with your data and query patterns suggest that non-default options will perform better in your specific case. - 4 bytes representing the current segment version as an integer. E.g., for v9 segments, the version is 0x0, 0x0, 0x0, 0x9 +For bitmap in string columns, the differences between using Roaring and Concise are most pronounced for high cardinality columns. In this case, Roaring is substantially faster on filters that match a lot of values, but in some cases Concise can have a lower footprint due to the overhead of the Roaring format (but is still slower when a lot of values are matched). You configure compression at the segment level, not for individual columns. See [IndexSpec](../ingestion/ingestion-spec.md#indexspec) for more details. -* `meta.smoosh` +## Segment identification - A file with metadata (filenames and offsets) about the contents of the other `smoosh` files +Segment identifiers typically contain the segment datasource, interval start time (in ISO 8601 format), interval end time (in ISO 8601 format), and version information. If data is additionally sharded beyond a time range, the segment identifier also contains a partition number: -* `XXXXX.smoosh` +`datasource_intervalStart_intervalEnd_version_partitionNum` - There are some number of these files, which are concatenated binary data +### Segment ID examples - The `smoosh` files represent multiple files "smooshed" together in order to minimize the number of file descriptors that must be open to house the data. They are files of up to 2GB in size (to match the limit of a memory mapped ByteBuffer in Java). The `smoosh` files house individual files for each of the columns in the data as well as an `index.drd` file with extra metadata about the segment. +The increasing partition numbers in the following segments indicate that multiple segments exist for the same interval: - There is also a special column called `__time` that refers to the time column of the segment. This will hopefully become less and less special as the code evolves, but for now it’s as special as my Mommy always told me I am. +``` +foo_2015-01-01/2015-01-02_v1_0 Review Comment: Logged as issue: https://github.com/apache/druid/issues/12614 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
