Pinot currently supports the following index techniques, where each of them have their own advantages in different query scenarios. By default, Pinot will use dictionary-encoded forward index for each column.

Forward index

Dictionary-encoded forward index with bit compression (default)

For each unique value from a column, we assign an id to it, and build a dictionary from the id to the value. Then in the forward index, we only store the bit-compressed ids instead of the values. With few number of unique values, dictionary-encoding can significantly improve the space efficiency of the storage.

The below diagram shows the dictionary encoding for two columns with integer and string types. As seen in the colA, dictionary encoding will save significant amount of space for duplicated values. On the other hand, colB has no duplicated data. Dictionary encoding will not compress much data in this case where there are a lot of unique values in the column. For string type, we pick the length of the longest value and use it as the length for dictionary’s fixed length value array. In this case, padding overhead can be high if there are a large number of unique values for a column.

Raw value forward index

In contrast to the dictionary-encoded forward index, raw value forward index directly stores values instead of ids.

Without the dictionary, the dictionary lookup step can be skipped for each value fetch. Also, the index can take advantage of the good locality of the values, thus improve the performance of scanning large number of values.

A typical use case to apply raw value forward index is when the column has a large number of unique values and the dictionary does not provide much compression. As seen the above diagram for dictionary encoding, scanning values with a dictionary involves a lot of random access because we need to perform dictionary look up. On the other hand, we can scan values sequentially with raw value forward index and this can improve performance a lot when applied appropriately.

Raw value forward index can be configured for a table by setting it in the table config as

{
    "tableIndexConfig": {
        "noDictionaryColumns": [
            "column_name",
            ...
        ],
        ...
    }
}

Sorted forward index with run-length encoding

When a column is physically sorted, Pinot uses a sorted forward index with run-length encoding on top of the dictionary-encoding. Instead of saving dictionary ids for each document id, we store a pair of start and end document id for each value. (The below diagram does not include dictionary encoding layer for simplicity.)

Sorted forward index has the advantages of both good compression and data locality. Sorted forward index can also be used as inverted index.

Sorted index can be configured for a table by setting it in the table config as

{
    "tableIndexConfig": {
        "sortedColumn": [
            "column_name"
        ],
        ...
    }
}

Real-time server will sort data on sortedColumn when generating segment internally. For offline push, input data needs to be sorted before running Pinot segment conversion and push job.

When applied correctly, one can find the following information on the segment metadata.

$ grep memberId <segment_name>/v3/metadata.properties | grep isSorted
column.memberId.isSorted = true

Inverted index

Bitmap inverted index

When inverted index is enabled for a column, Pinot maintains a map from each value to a bitmap, which makes value lookup to be constant time. When you have a column that is used for filtering frequently, adding an inverted index will improve the performance greatly.

Inverted index can be configured for a table by setting it in the table config as

{
    "tableIndexConfig": {
        "invertedIndexColumns": [
            "column_name",
            ...
        ],
        ...
    }
}

Sorted inverted index

Sorted forward index can directly be used as inverted index, with log(n) time lookup and it can benefit from data locality.

Sorted index performs much better than inverted index; however, it can only be applied to one column. When the query performance with inverted index is not good enough and most of queries have a filter on a specific column (e.g. memberId), sorted index can improve the query performance.

Star-tree index

Unlike other index techniques which work on single column, Star-Tree index is built on multiple columns, and utilize the pre-aggregated results to significantly reduce the number of values to be processed, thus improve the query performance.

One of the biggest challenges in realtime OLAP systems is achieving and maintaining tight SLA’s on latency and throughput on large data sets. Existing techniques such as sorted index or inverted index help improve query latencies, but speed-ups are still limited by number of documents necessary to process for computing the results. On the other hand, pre-aggregating the results ensures a constant upper bound on query latencies, but can lead to storage space explosion.

Here we introduce star-tree index to utilize the pre-aggregated documents in a smart way to achieve low query latencies but also use the storage space efficiently for aggregation/group-by queries.

Existing solutions

Consider the following data set as an example to discuss the existing approaches:

Sorted index

In this approach, data is sorted on a primary key, which is likely to appear as filter in most queries in the query set.

This reduces the time to search the documents for a given primary key value from linear scan O(n) to binary search O(logn), and also keeps good locality for the documents selected.

While this is a good improvement over linear scan, there are still a few issues with this approach:

• While sorting on one column does not require additional space, sorting on additional columns would require additional storage space to re-index the records for the various sort orders.

• While search time is reduced from O(n) to O(logn), overall latency is still a function of total number of documents need to be processed to answer a query.

Inverted index

In this approach, for each value of a given column, we maintain a list of document id’s where this value appears.

Below are the inverted indexes for columns ‘Browser’ and ‘Locale’ for our example data set:

For example, if we want to get all the documents where ‘Browser’ is ‘Firefox’, we can simply look up the inverted index for ‘Browser’ and identify that it appears in documents [1, 5, 6].

Using inverted index, we can reduce the search time to constant time O(1). However, the query latency is still a function of the selectivity of the query, i.e. increases with the number of documents need to be processed to answer the query.

Pre-aggregation

In this technique, we pre-compute the answer for a given query set upfront.

In the example below, we have pre-aggregated the total impressions for each country:

Doing so makes answering queries about total impressions for a country just a value lookup, by eliminating the need of processing a large number of documents. However, to be able to answer with multiple predicates implies pre-aggregating for various combinations of different dimensions. This leads to exponential explosion in storage space.

Star-tree solution

Space-Time Trade Off Between Different Techniques

We propose the Star-Tree data structure that offers a configurable trade-off between space and time and allows us to achieve hard upper bound for query latencies for a given use case. In the following sections we will define the Star-Tree data structure, and discuss how it is utilized within Pinot for achieving low latencies with high throughput.

Definitions

Tree Structure

Star-Tree Structure

• Root Node (Orange): Single root node, from which the rest of the tree can be traversed.

• Leaf Node (Blue): A leaf node can containing at most T records, where T is configurable.

• Non-leaf Node (Green): Nodes with more than T records are further split into children nodes.

• Star-Node (Yellow): Non-leaf nodes can also have a special child node called the Star-Node. This node contains the pre-aggregated records after removing the dimension on which the data was split for this level.

• Dimensions Split Order ([D1, D2]): Nodes at a given level in the tree are split into children nodes on all values of a particular dimension. The dimensions split order is an ordered list of dimensions that is used to determine the dimension to split on for a given level in the tree.

Node Properties

The properties stored in each node are as follows:

• Dimension: The dimension which the node is split on

• Start/End Document Id: The range of documents this node points to

• Aggregated Document Id: One single document which is the aggregation result of all documents pointed by this node

Index generation

Star-tree index is generated in the following steps:

• The data is first projected as per the dimensionsSplitOrder. Only the dimensions from the split order are reserved, others are dropped. For each unique combination of reserved dimensions, metrics are aggregated per configuration. The aggregated documents are written to a file and served as the initial Star-Tree documents (separate from the original documents).

• Sort the Star-Tree documents based on the dimensionsSplitOrder. It is primary-sorted on the first dimension in this list, and then secondary sorted on the rest of the dimensions based on their order in the list. Each node in the tree points to a range in the sorted documents.

• The tree structure can be created recursively (starting at root node) as follows:

  • If a node has more than T records, it is split into multiple children nodes, one for each value of the dimension in the split order corresponding to current level in the tree.

  • A Star-Node can be created (per configuration) for the current node, by dropping the dimension being split on, and aggregating the metrics for rows containing dimensions with identical values. These aggregated documents are appended to the end of the Star-Tree documents.

    If there is only one value for the current dimension, Star-Node won’t be created because the documents under the Star-Node are identical to the single node.

• The above step is repeated recursively until there are no more nodes to split.

• Multiple Star-Trees can be generated based on different configurations (dimensionsSplitOrder, aggregations, T)

Aggregation

Aggregation is configured as a pair of aggregation function and the column to apply the aggregation.

All types of aggregation function with bounded-sized intermediate result are supported.

Supported Functions

• COUNT

• MIN

• MAX

• SUM

• AVG

• MINMAXRANGE

• DISTINCTCOUNTHLL

• PERCENTILEEST

• PERCENTILETDIGEST

Unsupported Functions

• DISTINCTCOUNT: Intermediate result Set is unbounded

• PERCENTILE: Intermediate result List is unbounded

Index generation configuration

Multiple index generation configurations can be provided to generate multiple Star-Trees. Each configuration should contain the following properties:

• dimensionsSplitOrder: An ordered list of dimension names can be specified to configure the split order. Only the dimensions in this list are reserved in the aggregated documents. The nodes will be split based on the order of this list. For example, split at level i is performed on the values of dimension at index i in the list.

• skipStarNodeCreationForDimensions (Optional, default empty): A list of dimension names for which to not create the Star-Node.

• functionColumnPairs: A list of aggregation function and column pairs (split by double underscore “__”). E.g. SUM__Impressions (SUM of column Impressions)

• maxLeafRecords (Optional, default 10000): The threshold T to determine whether to further split each node.

Example

For our example data set, with the following example configuration, the tree and documents should be something like below.

"tableIndexConfig": {
  "starTreeIndexConfigs": [{
    "dimensionsSplitOrder": [
      "Country",
      "Browser",
      "Locale"
    ],
    "skipStarNodeCreationForDimensions": [
    ],
    "functionColumnPairs": [
      "SUM__Impressions"
    ],
    "maxLeafRecords": 1
  }],
  ...
}

Tree structure

The values in the parentheses are the aggregated sum of Impressions for all the documents under the node.

Star-tree documents

Query execution

For query execution, the idea is to first check metadata to determine whether the query can be solved with the Star-Tree documents, then traverse the Star-Tree to identify documents that satisfy all the predicates. After applying any remaining predicates that were missed while traversing the Star-Tree to the identified documents, apply aggregation/group-by on the qualified documents.

The algorithm to traverse the tree can be described as follows:

• Start from root node.

• For each level, what child node(s) to select depends on whether there are any predicates/group-by on the split dimension for the level in the query.

  • If there is no predicate or group-by on the split dimension, select the Star-Node if exists, or all child nodes to traverse further.

  • If there are predicate(s) on the split dimension, select the child node(s) that satisfy the predicate(s).

  • If there is no predicate, but there is a group-by on the split dimension, select all child nodes except Star-Node.

• Recursively repeat the previous step until all leaf nodes are reached, or all predicates are satisfied.

• Collect all the documents pointed to by the selected nodes.

  • If all predicates and group-bys are satisfied, pick the single aggregated document from each selected node.

  • Otherwise, collect all the documents in the document range from each selected node.

Notes on index tuning

If your use case is not site facing with a strict low latency requirement, inverted index will perform good enough for the most of use cases. We recommend to start with adding inverted index and if the query does not perform good enough, a user can consider to use more advanced indices such as sorted column and star-tree index.

Why do we need text search?

Pinot supports super fast query processing through its indexes on non-BLOB like columns. Queries with exact match filters are run efficiently through a combination of dictionary encoding, inverted index and sorted index. An example:

In the above query, we are doing exact match on two columns of type STRING and INT respectively.

In version 0.3.0, we added support for text indexes to efficiently do arbitrary search on STRING columns where each column value is a large BLOB of text. This can be achieved by using the new built-in function TEXT_MATCH.

where <column_name> is the column text index is created on and <search_expression> can be:

Sample Datasets

Text search should ideally be used on STRING columns where doing standard filter operations (EQUALITY, RANGE, BETWEEN) doesn't fit the bill because each column value is a reasonably large blob of text.

Apache Access Log

Consider the following snippet from Apache access log. Each line in the log consists of arbitrary data (IP addresses, URLs, timestamps, symbols etc) and represents a column value. Data like this is a good candidate for doing text search.

Let's say the following snippet of data is stored in ACCESS_LOG_COL column in Pinot table.

Few examples of search queries on this data:

Count the number of GET requests.

Count the number of POST requests that have administrator in the URL (administrator/index)

Count the number of POST requests that have a particular URL and handled by Firefox browser

Resume text

Consider another example of simple resume text. Each line in the file represents skill-data from resumes of different candidates

Let's say the following snippet of data is stored in SKILLS_COL column in Pinot table. Each line in the input text represents a column value.

Few examples of search queries on this data:

Count the number of candidates that have "machine learning" and "gpu processing" - a phrase search (more on this further in the document) where we are looking for exact match of phrases "machine learning" and "gpu processing" not necessarily in the same order in original data.

Count the number of candidates that have "distributed systems" and either 'Java' or 'C++' - a combination of searching for exact phrase "distributed systems" along with other terms.

Query Log

Consider a snippet from a log file containing SQL queries handled by a database. Each line (query) in the file represents a column value in QUERY_LOG_COL column in Pinot table.

Few examples of search queries on this data:

Count the number of queries that have GROUP BY

Count the number of queries that have the SELECT count... pattern

Count the number of queries that use BETWEEN filter on timestamp column along with GROUP BY

Current restrictions

Currently we support text search in a restricted manner. More specifically, we have the following constraints:

• The column type should be STRING.

• The column should be single-valued.

• Co-existence of text index with other Pinot indexes is currently not supported.

The last two restrictions are going to be relaxed very soon in the upcoming releases.

Co-existence with other indexes

Currently, a column in Pinot can be dictionary encoded or stored RAW. Furthermore, we can create inverted index on the dictionary encoded column. We can also create a sorted index on the dictionary encoded column.

Text index is an addition to the type of per-column indexes users can create in Pinot. However, the current implementation supports text index on RAW column. In other words, the column should not be dictionary encoded. As we relax this constraint in upcoming releases, text index can be created on a dictionary encoded column that also has other indexes (inverted, sorted etc).

How to enable text index?

Similar to other indexes, users can enable text index on a column through table config. As part of text-search feature, we have also introduced a new generic way of specifying the per-column encoding and index information. In the table config, there will be a new section with name "fieldConfigList".

IMPORTANT: This mechanism of using "fieldConfigList" is currently ONLY used for text indexes. Our plan is to migrate all other indexes to this model. We are going to do that in upcoming releases and accordingly user documentation and new guidelines will be published. So please continue to specify other index info in table config as you have done till now and use the "fieldConfigList" only for text indexes.

"fieldConfigList" will be a new section in table config. It is essentially a list of per-column encoding and index information. In the above example, the list contains text index information for two columns text_col_1 and text_col_2. Each object in fieldConfigList contains the following information

• name - Name of the column text index is enabled on

• encodingType - As mentioned earlier, we can store a column either as RAW or dictionary encoded. Since for now we have a restriction on the text index, this should always be RAW.

• indexType - This should be TEXT.

Also, since we haven't yet removed the old way of specifying the index info, each column that text index is enabled on should also be specified in noDictionaryColumns in tableIndexConfig

The above mechanism should allow the user to use text index in all of the following scenarios:

• Adding new table with text index enabled on one or more columns.

• Adding a new column with text index enabled to an existing table.

• Enabling text index on an existing column.

Since we haven't yet removed the old way of specifying the

Text Index Creation

Once the text index is enabled on one or more columns through table config, our segment generation code will pick up the config and automatically create text index (per column). This is exactly how other indexes in Pinot are created.

Text index is supported for both offline and realtime segments.

Text parsing and tokenization

The original text document (a value in the column with text index enabled) is parsed, tokenized and individual "indexable" terms are extracted. These terms are inserted into the index.

Pinot's text index is built on top of Lucene. Lucene's standard english text tokenizer generally works well for most classes of text. We might want to build custom text parser and tokenizer to suit particular user requirements. Accordingly, we can make this configurable for the user to specify on per column text index basis.

Writing Text Search Queries

A new built-in function TEXT_MATCH has been introduced for using text search in SQL/PQL.

TEXT_MATCH(text_column_name, search_expression)

• text_column_name - name of the column to do text search on.

• search_expression - search query

We can use TEXT_MATCH function as part of our queries in the WHERE clause. Examples:

We can also use the TEXT_MATCH filter clause with other filter operators. For example:

Combining multiple TEXT_MATCH filter clauses

TEXT_MATCH can be used in WHERE clause of all kinds of queries supported by Pinot

• Selection query which projects one or more columns

  • User can also include the text column name in select list

• Aggregation query

• Aggregation GROUP BY query

The search expression (second argument to TEXT_MATCH function) is the query string that Pinot will use to perform text search on the column's text index. **Following expression types are supported

Phrase Query

This query is used to do exact match of a given phrase. Exact match implies that terms in the user specified phrase should appear in the exact same order in the original text document. Note that document is referred to as the column value.

Let's take the example of resume text data containing 14 documents to walk through queries. The data is stored in column named SKILLS_COL and we have created a text index on this column.

Example 1 - Search in SKILL_COL column to look for documents where each matching document MUST contain phrase "distributed systems" as is

The search expression is '\"Distributed systems\"'

• The search expression is always specified within single quotes '<your expression>'

• Since we are doing a phrase search, the phrase should be specified within double quotes inside the single quotes and the double quotes should be escaped

  • '\"<your phrase>\"'

The above query will match the following documents:

But it won't match the following document:

This is because the phrase query looks for the phrase occurring in the original document "as is". The terms as specified by the user in phrase should be in the exact same order in the original document for the document to be considered as a match.

NOTE: Matching is always done in a case-insensitive manner.

Example 2 - Search in SKILL_COL column to look for documents where each matching document MUST contain phrase "query processing" as is

The above query will match the following documents:

Term Query

Term queries are used to search for individual terms

Example 3 - Search in SKILL_COL column to look for documents where each matching document MUST contain the term 'java'

As mentioned earlier, the search expression is always within single quotes. However, since this is a term query, we don't have to use double quotes within single quotes.

Composite Query using Boolean Operators

Boolean operators AND, OR are supported and we can use them to build a composite query. Boolean operators can be used to combine phrase and term queries in any arbitrary manner

Example 4 - Search in SKILL_COL column to look for documents where each matching document MUST contain phrases "distributed systems" and "tensor flow". This combines two phrases using AND boolean operator

The above query will match the following documents:

Example 5 - Search in SKILL_COL column to look for documents where each document MUST contain phrase "machine learning" and term 'gpu' and term 'python'. This combines a phrase and two terms using boolean operator

The above query will match the following documents:

When using boolean operators to combine term(s) and phrase(s) or both, please note that:

• The matching document can contain the terms and phrases in any order.

• The matching document may not have the terms adjacent to each other (if this is needed, please use appropriate phrase query for the concerned terms).

Use of OR operator is implicit. In other words, if phrase(s) and term(s) are not combined using AND operator in the search expression, OR operator is used by default:

Example 6 - Search in SKILL_COL column to look for documents where each document MUST contain ANY one of:

• phrase "distributed systems" OR

• term 'java' OR

• term 'C++'.

We can also do grouping using parentheses:

Example 7 - Search in SKILL_COL column to look for documents where each document MUST contain

• phrase "distributed systems" AND

• at least one of the terms Java or C++

In the below query, we group terms Java and C++ without any operator which implies the use of OR. The root operator AND is used to combine this with phrase "distributed systems"

Prefix Query

Prefix searches can also be done in the context of a single term. We can't use prefix matches for phrases.

Example 8 - Search in SKILL_COL column to look for documents where each document MUST contain text like stream, streaming, streams etc

The above query will match the following documents:

Regular Expression Query

Phrase and term queries work on the fundamental logic of looking up the terms (aka tokens) in the text index. The original text document (a value in the column with text index enabled) is parsed, tokenized and individual "indexable" terms are extracted. These terms are inserted into the index.

Based on the nature of original text and how the text is segmented into tokens, it is possible that some terms don't get indexed individually. In such cases, it is better to use regular expression queries on the text index.

Consider server log as an example and we want to look for exceptions. A regex query is suitable for this scenario as it is unlikely that 'exception' is present as an individual indexed token.

Syntax of a regex query is slightly different from queries mentioned earlier. The regular expression is written between a pair of forward slashes (/).

The above query will match any text document containing exception.

Deciding Query Types

Generally, a combination of phrase and term queries using boolean operators and grouping should allow us to build a complex text search query expression.

The key thing to remember is that phrases should be used when the order of terms in the document is important and if separating the phrase into individual terms doesn't make sense from end user's perspective.

An example would be phrase "machine learning".

However, if we are searching for documents matching Java and C++ terms, using phrase query "Java C++" will actually result in in partial results (could be empty too) since now we are relying the on the user specifying these skills in the exact same order (adjacent to each other) in the resume text.

Term query using boolean AND operator is more appropriate for such cases

Now that the QuickStartCluster is setup, we can start exploring the data and the APIs. Head over to http://localhost:9000 in your browser.

You are now connected to the Pinot controller. Let's take a look at the following two features.

Query Console

Query Console let's us run queries on the data in the Pinot cluster

We can see our baseballStats table listed on the left (you will see meetupRSVP or airlineStats if you used the streaming or the hybrid quick start). Clicking on the table name should display all the names and data types of the columns of the table, and also execute a sample query select * from baseballStats limit 10 . You can query this table by typing your query in the text box and clicking the Run Query button.

Here's some other queries you can try out:

select playerName, max(hits) from baseballStats group by playerName order by max(hits) desc

select sum(hits), sum(homeRuns), sum(numberOfGames) from baseballStats where yearID > 2010

select * from baseballStats order by league

Pinot supports a subset of standard SQL. See Pinot Query Language for more information.

Rest API

The Pinot Admin UI contains all the APIs that you will need to operate and manage your cluster. It provides a set of APIs for Pinot cluster management including health check, instances management, schema and table management, data segments management.

Let's check out the tables in this cluster by going to Table -> List all tables in cluster and click on Try it out!. We can see the baseballStats table listed here. We can also see the exact curl call made to the controller API.

You can look at the configuration of this table by going to Tables -> Get/Enable/Disable/Drop a table, type in baseballStats in the table name, and click Try it out!

Let's check out the schemas in the cluster by going to Schema -> List all schemas in the cluster and click Try it out!. We can see a schema called baseballStats in this list.

Take a look at the schema by going to Schema -> Get a schema, type baseballStats in the schema name, and click Try it out!.

{
  "schemaName": "baseballStats",
  "dimensionFieldSpecs": [
    {
      "name": "playerID",
      "dataType": "STRING"
    },
    {
      "name": "yearID",
      "dataType": "INT"
    },
    {
      "name": "teamID",
      "dataType": "STRING"
    },
    {
      "name": "league",
      "dataType": "STRING"
    },
    {
      "name": "playerName",
      "dataType": "STRING"
    }
  ],
  "metricFieldSpecs": [
    {
      "name": "playerStint",
      "dataType": "INT"
    },
    {
      "name": "numberOfGames",
      "dataType": "INT"
    },
    {
      "name": "numberOfGamesAsBatter",
      "dataType": "INT"
    },
    {
      "name": "AtBatting",
      "dataType": "INT"
    },
    {
      "name": "runs",
      "dataType": "INT"
    },
    {
      "name": "hits",
      "dataType": "INT"
    },
    {
      "name": "doules",
      "dataType": "INT"
    },
    {
      "name": "tripples",
      "dataType": "INT"
    },
    {
      "name": "homeRuns",
      "dataType": "INT"
    },
    {
      "name": "runsBattedIn",
      "dataType": "INT"
    },
    {
      "name": "stolenBases",
      "dataType": "INT"
    },
    {
      "name": "caughtStealing",
      "dataType": "INT"
    },
    {
      "name": "baseOnBalls",
      "dataType": "INT"
    },
    {
      "name": "strikeouts",
      "dataType": "INT"
    },
    {
      "name": "intentionalWalks",
      "dataType": "INT"
    },
    {
      "name": "hitsByPitch",
      "dataType": "INT"
    },
    {
      "name": "sacrificeHits",
      "dataType": "INT"
    },
    {
      "name": "sacrificeFlies",
      "dataType": "INT"
    },
    {
      "name": "groundedIntoDoublePlays",
      "dataType": "INT"
    },
    {
      "name": "G_old",
      "dataType": "INT"
    }
  ]
}

Finally, let's checkout the data segments in the cluster by going to Segment -> List all segments, type in baseballStats in the table name, and click Try it out!. There's 1 segment for this table, called baseballStats_OFFLINE_0.

You might have figured out by now, in order to get data into the Pinot cluster, we need a table, a schema and segments. Let's head over to Batch upload sample data, to find out more about these components and learn how to create them for your own data.