Multi-queries

Multi-queries, or query batches, allow you to send multiple search queries to Manticore in a single network request.

👍 Why use multi-queries?

The primary reason is performance. By sending requests to Manticore in a batch instead of one by one, you save time by reducing network round-trips. Additionally, sending queries in a batch allows Manticore to perform certain internal optimizations. If no batch optimizations can be applied, queries will be processed individually.

⛔ When not to use multi-queries?

Multi-queries require all search queries in a batch to be independent, which isn't always the case. Sometimes query B depends on query A's results, meaning query B can only be set up after executing query A. For example, you might want to display results from a secondary index only if no results were found in the primary table, or you may want to specify an offset into the 2nd result set based on the number of matches in the 1st result set. In these cases, you'll need to use separate queries (or separate batches).

You can run multiple search queries with SQL by separating them with a semicolon. When Manticore receives a query formatted like this from a client, all inter-statement optimizations will be applied.

Multi-queries don't support queries with FACET. The number of multi-queries in one batch shouldn't exceed max_batch_queries.

‹›
  • SQL
SQL
📋
SELECT id, price FROM products WHERE MATCH('remove hair') ORDER BY price DESC; SELECT id, price FROM products WHERE MATCH('remove hair') ORDER BY price ASC

Multi-queries optimizations

There are two major optimizations to be aware of: common query optimization and common subtree optimization.

Common query optimization means that searchd will identify all those queries in a batch where only the sorting and group-by settings differ, and only perform searching once. For example, if a batch consists of 3 queries, all of them are for "ipod nano", but the 1st query requests the top-10 results sorted by price, the 2nd query groups by vendor ID and requests the top-5 vendors sorted by rating, and the 3rd query requests the max price, full-text search for "ipod nano" will only be performed once, and its results will be reused to build 3 different result sets.

Faceted search is a particularly important case that benefits from this optimization. Indeed, faceted searching can be implemented by running several queries, one to retrieve search results themselves, and a few others with the same full-text query but different group-by settings to retrieve all the required groups of results (top-3 authors, top-5 vendors, etc). As long as the full-text query and filtering settings stay the same, common query optimization will trigger, and greatly improve performance.

Common subtree optimization is even more interesting. It allows searchd to exploit similarities between batched full-text queries. It identifies common full-text query parts (subtrees) in all queries and caches them between queries. For example, consider the following query batch:

donald trump president
donald trump barack obama john mccain
donald trump speech

There's a common two-word part donald trump that can be computed only once, then cached and shared across the queries. And common subtree optimization does just that. Per-query cache size is strictly controlled by subtree_docs_cache and subtree_hits_cache directives (so that caching all sixteen gazillions of documents that match "i am" does not exhaust the RAM and instantly kill your server).

How can you tell if the queries in the batch were actually optimized? If they were, the respective query log will have a "multiplier" field that specifies how many queries were processed together:

Note the "x3" field. It means that this query was optimized and processed in a sub-batch of 3 queries.

‹›
  • log
log
📋
[Sun Jul 12 15:18:17.000 2009] 0.040 sec x3 [ext/0/rel 747541 (0,20)] [lj] the
[Sun Jul 12 15:18:17.000 2009] 0.040 sec x3 [ext/0/ext 747541 (0,20)] [lj] the
[Sun Jul 12 15:18:17.000 2009] 0.040 sec x3 [ext/0/ext 747541 (0,20)] [lj] the

For reference, this is how the regular log would look like if the queries were not batched:

‹›
  • log
log
📋
[Sun Jul 12 15:18:17.062 2009] 0.059 sec [ext/0/rel 747541 (0,20)] [lj] the
[Sun Jul 12 15:18:17.156 2009] 0.091 sec [ext/0/ext 747541 (0,20)] [lj] the
[Sun Jul 12 15:18:17.250 2009] 0.092 sec [ext/0/ext 747541 (0,20)] [lj] the

Notice how the per-query time in the multi-query case improved by a factor of 1.5x to 2.3x, depending on the specific sorting mode.

Sub-selects

Manticore supports SELECT subqueries via SQL in the following format:

SELECT * FROM (SELECT ... ORDER BY cond1 LIMIT X) ORDER BY cond2 LIMIT Y

The outer select allows only ORDER BY and LIMIT clauses. Sub-select queries currently have two use cases:

  1. When you have a query with two ranking UDFs, one very fast and the other slow, and perform a full-text search with a large match result set. Without subselect, the query would look like:

     SELECT id,slow_rank() as slow,fast_rank() as fast FROM index
         WHERE MATCH(‘some common query terms’) ORDER BY fast DESC, slow DESC LIMIT 20
         OPTION max_matches=1000;

    With sub-selects, the query can be rewritten as:

     SELECT * FROM
         (SELECT id,slow_rank() as slow,fast_rank() as fast FROM index WHERE
             MATCH(‘some common query terms’)
             ORDER BY fast DESC LIMIT 100 OPTION max_matches=1000)
     ORDER BY slow DESC LIMIT 20;

    In the initial query, the slow_rank() UDF is computed for the entire match result set. With SELECT sub-queries, only fast_rank() is computed for the entire match result set, while slow_rank() is computed for a limited set.

  2. The second case is useful for large result sets coming from a distributed table.

    For this query:

     SELECT * FROM my_dist_index WHERE some_conditions LIMIT 50000;

    If you have 20 nodes, each node can send back to the master a maximum of 50K records, resulting in 20 x 50K = 1M records. However, since the master sends back only 50K (out of 1M), it might be good enough for the nodes to send only the top 10K records. With sub-select, you can rewrite the query as:

     SELECT * FROM
          (SELECT * FROM my_dist_index WHERE some_conditions LIMIT 10000)
      ORDER by some_attr LIMIT 50000;

    In this case, the nodes receive only the inner query and execute it. This means the master will receive only 20x10K=200K records. The master will take all the records received, reorder them by the OUTER clause, and return the best 50K records. The sub-select helps reduce the traffic between the master and the nodes, as well as reduce the master's computation time (since it processes only 200K instead of 1M records).

Grouping search results

Grouping search results is often helpful for obtaining per-group match counts or other aggregations. For example, it's useful for creating a graph illustrating the number of matching blog posts per month or grouping web search results by site or forum posts by author, etc.

Manticore supports the grouping of search results by single or multiple columns and computed expressions. The results can:

  • Be sorted within a group
  • Return more than one row per group
  • Have groups filtered
  • Have groups sorted
  • Be aggregated using the aggregation functions
‹›
  • SQL
  • JSON
📋

General syntax

SELECT {* | SELECT_expr [, SELECT_expr ...]}
...
GROUP BY {field_name | alias } [, ...]
[HAVING where_condition]
[WITHIN GROUP ORDER BY field_name {ASC | DESC} [, ...]]
...

SELECT_expr: { field_name | function_name(...) }
where_condition: {aggregation expression alias | COUNT(*)}

Just Grouping

Grouping is quite simple - just add "GROUP BY smth" to the end of your SELECT query. The something can be:

  • Any non-full-text field from the table: integer, float, string, MVA (multi-value attribute)
  • Or, if you used an alias in the SELECT list, you can GROUP BY it too

You can omit any aggregation functions in the SELECT list and it will still work:

‹›
  • SQL
SQL
📋
SELECT release_year FROM films GROUP BY release_year LIMIT 5;
‹›
Response
+--------------+
| release_year |
+--------------+
|         2004 |
|         2002 |
|         2001 |
|         2005 |
|         2000 |
+--------------+

In most cases, however, you'll want to obtain some aggregated data for each group, such as:

  • COUNT(*) to simply get the number of elements in each group
  • or AVG(field) to calculate the average value of the field within the group
‹›
  • SQL1
  • SQL2
  • JSON
  • PHP
  • Python
  • Javascript
  • Java
  • C#
📋
SELECT release_year, count(*) FROM films GROUP BY release_year LIMIT 5;
‹›
Response
+--------------+----------+
| release_year | count(*) |
+--------------+----------+
|         2004 |      108 |
|         2002 |      108 |
|         2001 |       91 |
|         2005 |       93 |
|         2000 |       97 |
+--------------+----------+
Sorting groups

By default, groups are not sorted, and the next thing you typically want to do is order them by something, like the field you're grouping by:

‹›
  • SQL
SQL
📋
SELECT release_year, count(*) from films GROUP BY release_year ORDER BY release_year asc limit 5;
‹›
Response
+--------------+----------+
| release_year | count(*) |
+--------------+----------+
|         2000 |       97 |
|         2001 |       91 |
|         2002 |      108 |
|         2003 |      106 |
|         2004 |      108 |
+--------------+----------+

Alternatively, you can sort by the aggregation:

  • by count(*) to display groups with the most elements first
  • by avg(rental_rate) to show the highest-rated movies first. Note that in the example, it's done via an alias: avg(rental_rate) is first mapped to avg in the SELECT list, and then we simply do ORDER BY avg
‹›
  • SQL1
  • SQL2
📋
SELECT release_year, count(*) FROM films GROUP BY release_year ORDER BY count(*) desc LIMIT 5;
‹›
Response
+--------------+----------+
| release_year | count(*) |
+--------------+----------+
|         2004 |      108 |
|         2002 |      108 |
|         2003 |      106 |
|         2006 |      103 |
|         2008 |      102 |
+--------------+----------+
GROUP BY multiple fields at once

In some cases, you might want to group not just by a single field, but by multiple fields at once, such as a movie's category and year:

‹›
  • SQL
SQL
📋
SELECT category_id, release_year, count(*) FROM films GROUP BY category_id, release_year ORDER BY category_id ASC, release_year ASC;
‹›
Response
+-------------+--------------+----------+
| category_id | release_year | count(*) |
+-------------+--------------+----------+
|           1 |         2000 |        5 |
|           1 |         2001 |        2 |
|           1 |         2002 |        6 |
|           1 |         2003 |        6 |
|           1 |         2004 |        5 |
|           1 |         2005 |       10 |
|           1 |         2006 |        4 |
|           1 |         2007 |        5 |
|           1 |         2008 |        7 |
|           1 |         2009 |       14 |
|           2 |         2000 |       10 |
|           2 |         2001 |        5 |
|           2 |         2002 |        6 |
|           2 |         2003 |        6 |
|           2 |         2004 |       10 |
|           2 |         2005 |        4 |
|           2 |         2006 |        5 |
|           2 |         2007 |        8 |
|           2 |         2008 |        8 |
|           2 |         2009 |        4 |
+-------------+--------------+----------+
Give me N rows

Sometimes it's useful to see not just a single element per group, but multiple. This can be easily achieved with the help of GROUP N BY. For example, in the following case, we get two movies for each year rather than just one, which a simple GROUP BY release_year would have returned.

‹›
  • SQL
SQL
📋
SELECT release_year, title FROM films GROUP 2 BY release_year ORDER BY release_year DESC LIMIT 6;
‹›
Response
+--------------+-----------------------------+
| release_year | title                       |
+--------------+-----------------------------+
|         2009 | ALICE FANTASIA              |
|         2009 | ALIEN CENTER                |
|         2008 | AMADEUS HOLY                |
|         2008 | ANACONDA CONFESSIONS        |
|         2007 | ANGELS LIFE                 |
|         2007 | ARACHNOPHOBIA ROLLERCOASTER |
+--------------+-----------------------------+
Sorting inside a group

Another crucial analytics requirement is to sort elements within a group. To achieve this, use the WITHIN GROUP ORDER BY ... {ASC|DESC} clause. For example, let's get the highest-rated film for each year. Note that it works in parallel with just ORDER BY:

  • WITHIN GROUP ORDER BY sorts results inside a group
  • while just GROUP BY sorts the groups themselves

These two work entirely independently.

‹›
  • SQL
SQL
📋
SELECT release_year, title, rental_rate FROM films GROUP BY release_year WITHIN GROUP ORDER BY rental_rate DESC ORDER BY release_year DESC LIMIT 5;
‹›
Response
+--------------+------------------+-------------+
| release_year | title            | rental_rate |
+--------------+------------------+-------------+
|         2009 | AMERICAN CIRCUS  |    4.990000 |
|         2008 | ANTHEM LUKE      |    4.990000 |
|         2007 | ATTACKS HATE     |    4.990000 |
|         2006 | ALADDIN CALENDAR |    4.990000 |
|         2005 | AIRPLANE SIERRA  |    4.990000 |
+--------------+------------------+-------------+
Filter groups

HAVING expression is a helpful clause for filtering groups. While WHERE is applied before grouping, HAVING works with the groups. For example, let's keep only those years when the average rental rate of the films for that year was higher than 3. We get only four years:

‹›
  • SQL
SQL
📋
SELECT release_year, avg(rental_rate) avg FROM films GROUP BY release_year HAVING avg > 3;
‹›
Response
+--------------+------------+
| release_year | avg        |
+--------------+------------+
|         2002 | 3.08259249 |
|         2001 | 3.09989142 |
|         2000 | 3.17556739 |
|         2006 | 3.26184368 |
+--------------+------------+
GROUPBY()

There is a function GROUPBY() which returns the key of the current group. It's useful in many cases, especially when you GROUP BY an MVA or a JSON value.

It can also be used in HAVING, for example, to keep only years 2000 and 2002.

Note that GROUPBY()is not recommended for use when you GROUP BY multiple fields at once. It will still work, but since the group key in this case is a compound of field values, it may not appear the way you expect.

‹›
  • SQL
SQL
📋
SELECT release_year, count(*) FROM films GROUP BY release_year HAVING GROUPBY() IN (2000, 2002);
‹›
Response
+--------------+----------+
| release_year | count(*) |
+--------------+----------+
|         2002 |      108 |
|         2000 |       97 |
+--------------+----------+
Grouping by MVA (multi-value attributes)

Manticore supports grouping by MVA. To demonstrate how it works, let's create a table "shoes" with MVA "sizes" and insert a few documents into it:

create table shoes(title text, sizes multi);
insert into shoes values(0,'nike',(40,41,42)),(0,'adidas',(41,43)),(0,'reebook',(42,43));

so we have:

SELECT * FROM shoes;
+---------------------+----------+---------+
| id                  | sizes    | title   |
+---------------------+----------+---------+
| 1657851069130080265 | 40,41,42 | nike    |
| 1657851069130080266 | 41,43    | adidas  |
| 1657851069130080267 | 42,43    | reebook |
+---------------------+----------+---------+

If we now GROUP BY "sizes", it will process all our multi-value attributes and return an aggregation for each, in this case just the count:

‹›
  • SQL
  • JSON
  • PHP
  • Python
  • Javascript
  • Java
  • C#
📋
SELECT groupby() gb, count(*) FROM shoes GROUP BY sizes ORDER BY gb asc;
‹›
Response
+------+----------+
| gb   | count(*) |
+------+----------+
|   40 |        1 |
|   41 |        2 |
|   42 |        2 |
|   43 |        2 |
+------+----------+
Grouping by a JSON node

If you have a field of type JSON, you can GROUP BY any node from it. To demonstrate this, let's create a table "products" with a few documents, each having a color in the "meta" JSON field:

create table products(title text, meta json);
insert into products values(0,'nike','{"color":"red"}'),(0,'adidas','{"color":"red"}'),(0,'puma','{"color":"green"}');

This gives us:

SELECT * FROM products;
+---------------------+-------------------+--------+
| id                  | meta              | title  |
+---------------------+-------------------+--------+
| 1657851069130080268 | {"color":"red"}   | nike   |
| 1657851069130080269 | {"color":"red"}   | adidas |
| 1657851069130080270 | {"color":"green"} | puma   |
+---------------------+-------------------+--------+

To group the products by color, we can simply use GROUP BY meta.color, and to display the corresponding group key in the SELECT list, we can use GROUPBY():

‹›
  • SQL
  • JSON
  • PHP
  • Python
  • Javascript
  • Java
  • C#
📋
SELECT groupby() color, count(*) from products GROUP BY meta.color;
‹›
Response
+-------+----------+
| color | count(*) |
+-------+----------+
| red   |        2 |
| green |        1 |
+-------+----------+

Aggregation functions

Besides COUNT(*), which returns the number of elements in each group, you can use various other aggregation functions:

COUNT(DISTINCT field)

While COUNT(*) returns the number of all elements in the group, COUNT(DISTINCT field) returns the number of unique values of the field in the group, which may be completely different from the total count. For instance, you can have 100 elements in the group, but all with the same value for a certain field. COUNT(DISTINCT field) helps to determine that. To demonstrate this, let's create a table "students" with the student's name, age, and major:

CREATE TABLE students(name text, age int, major string);
INSERT INTO students values(0,'John',21,'arts'),(0,'William',22,'business'),(0,'Richard',21,'cs'),(0,'Rebecca',22,'cs'),(0,'Monica',21,'arts');

so we have:

MySQL [(none)]> SELECT * from students;
+---------------------+------+----------+---------+
| id                  | age  | major    | name    |
+---------------------+------+----------+---------+
| 1657851069130080271 |   21 | arts     | John    |
| 1657851069130080272 |   22 | business | William |
| 1657851069130080273 |   21 | cs       | Richard |
| 1657851069130080274 |   22 | cs       | Rebecca |
| 1657851069130080275 |   21 | arts     | Monica  |
+---------------------+------+----------+---------+

In the example, you can see that if we GROUP BY major and display both COUNT(*) and COUNT(DISTINCT age), it becomes clear that there are two students who chose the major "cs" with two unique ages, but for the major "arts", there are also two students, yet only one unique age.

There can be at most one COUNT(DISTINCT) per query.

By default, counts are approximate

Actually, some of them are exact, while others are approximate. More on that below.

Manticore supports two algorithms for computing counts of distinct values. One is a legacy algorithm that uses a lot of memory and is usually slow. It collects {group; value} pairs, sorts them, and periodically discards duplicates. The benefit of this approach is that it guarantees exact counts within a plain index. You can enable it by setting the distinct_precision_threshold option to 0.

The other algorithm (enabled by default) loads counts into a hash table and returns its size. If the hash table becomes too large, its contents are moved into a HyperLogLog. This is where the counts become approximate since HyperLogLog is a probabilistic algorithm. The advantage is that the maximum memory usage per group is fixed and depends on the accuracy of the HyperLogLog. The overall memory usage also depends on the max_matches setting, which reflects the number of groups.

The distinct_precision_threshold option sets the threshold below which counts are guaranteed to be exact. The HyperLogLog accuracy setting and the threshold for the "hash table to HyperLogLog" conversion are derived from this setting. It's important to use this option with caution because doubling it will double the maximum memory required for count calculations. The maximum memory usage can be roughly estimated using this formula: 64 * max_matches * distinct_precision_threshold. Note that this is the worst-case scenario, and in most cases, count calculations will use significantly less RAM.

COUNT(DISTINCT) against a distributed table or a real-time table consisting of multiple disk chunks may return inaccurate results, but the result should be accurate for a distributed table consisting of local plain or real-time tables with the same schema (identical set/order of fields, but may have different tokenization settings).

‹›
  • SQL
SQL
📋
SELECT major, count(*), count(distinct age) FROM students GROUP BY major;
‹›
Response
+----------+----------+---------------------+
| major    | count(*) | count(distinct age) |
+----------+----------+---------------------+
| arts     |        2 |                   1 |
| business |        1 |                   1 |
| cs       |        2 |                   2 |
+----------+----------+---------------------+
GROUP_CONCAT(field)

Often, you want to better understand the contents of each group. You can use GROUP N BY for that, but it would return additional rows you might not want in the output. GROUP_CONCAT() enriches your grouping by concatenating values of a specific field in the group. Let's take the previous example and improve it by displaying all the ages in each group.

GROUP_CONCAT(field) returns the list as comma-separated values.

‹›
  • SQL
SQL
📋
SELECT major, count(*), count(distinct age), group_concat(age) FROM students GROUP BY major
‹›
Response
+----------+----------+---------------------+-------------------+
| major    | count(*) | count(distinct age) | group_concat(age) |
+----------+----------+---------------------+-------------------+
| arts     |        2 |                   1 | 21,21             |
| business |        1 |                   1 | 22                |
| cs       |        2 |                   2 | 21,22             |
+----------+----------+---------------------+-------------------+
SUM(), MIN(), MAX(), AVG()

Of course, you can also obtain the sum, average, minimum, and maximum values within a group.

‹›
  • SQL
SQL
📋
SELECT release_year year, sum(rental_rate) sum, min(rental_rate) min, max(rental_rate) max, avg(rental_rate) avg FROM films GROUP BY release_year ORDER BY year asc LIMIT 5;
‹›
Response
+------+------------+----------+----------+------------+
| year | sum        | min      | max      | avg        |
+------+------------+----------+----------+------------+
| 2000 | 308.030029 | 0.990000 | 4.990000 | 3.17556739 |
| 2001 | 282.090118 | 0.990000 | 4.990000 | 3.09989142 |
| 2002 | 332.919983 | 0.990000 | 4.990000 | 3.08259249 |
| 2003 | 310.940063 | 0.990000 | 4.990000 | 2.93339682 |
| 2004 | 300.920044 | 0.990000 | 4.990000 | 2.78629661 |
+------+------------+----------+----------+------------+

Grouping accuracy

Grouping is done in fixed memory, which depends on the max_matches setting. If max_matches allows for storage of all found groups, the results will be 100% accurate. However, if the value of max_matches is lower, the results will be less accurate.

When parallel processing is involved, it can become more complicated. When pseudo_sharding is enabled and/or when using an RT index with several disk chunks, each chunk or pseudo shard gets a result set that is no larger than max_matches. This can lead to inaccuracies in aggregates and group counts when the result sets from different threads are merged. To fix this, either a larger max_matches value or disabling parallel processing can be used.

Manticore will try to increase max_matches up to max_matches_increase_threshold if it detects that groupby may return inaccurate results. Detection is based on the number of unique values of the groupby attribute, which is retrieved from secondary indexes (if present).

To ensure accurate aggregates and/or group counts when using RT indexes or pseudo_sharding, accurate_aggregation can be enabled. This will try to increase max_matches up to the threshold, and if the threshold is not high enough, Manticore will disable parallel processing for the query.

‹›
  • SQL
SQL
📋
MySQL [(none)]> SELECT release_year year, count(*) FROM films GROUP BY year limit 5;
+------+----------+
| year | count(*) |
+------+----------+
| 2004 |      108 |
| 2002 |      108 |
| 2001 |       91 |
| 2005 |       93 |
| 2000 |       97 |
+------+----------+

MySQL [(none)]> SELECT release_year year, count(*) FROM films GROUP BY year limit 5 option max_matches=1;
+------+----------+
| year | count(*) |
+------+----------+
| 2004 |       76 |
+------+----------+

MySQL [(none)]> SELECT release_year year, count(*) FROM films GROUP BY year limit 5 option max_matches=2;
+------+----------+
| year | count(*) |
+------+----------+
| 2004 |       76 |
| 2002 |       74 |
+------+----------+

MySQL [(none)]> SELECT release_year year, count(*) FROM films GROUP BY year limit 5 option max_matches=3;
+------+----------+
| year | count(*) |
+------+----------+
| 2004 |      108 |
| 2002 |      108 |
| 2001 |       91 |
+------+----------+