Main+delta schema

In many situations, the total dataset is too large to be frequently rebuilt from scratch, while the number of new records remains relatively small. For example, a forum may have 1,000,000 archived posts but only receive 1,000 new posts per day.

In such cases, implementing "live" (nearly real-time) table updates can be achieved using a "main+delta" scheme.

The concept involves setting up two sources and two tables, with one "main" table for data that rarely changes (if ever), and one "delta" table for new documents. In the example, the 1,000,000 archived posts would be stored in the main table, while the 1,000 new daily posts would be placed in the delta table. The delta table can then be rebuilt frequently, making the documents available for searching within seconds or minutes. Determining which documents belong to which table and rebuilding the main table can be fully automated. One approach is to create a counter table that tracks the ID used to split the documents and update it whenever the main table is rebuilt.

Using a timestamp column as the split variable is more effective than using the ID since timestamps can track not only new documents but also modified ones.

For datasets that may contain modified or deleted documents, the delta table should provide a list of affected documents, ensuring they are suppressed and excluded from search queries. This is accomplished using a feature called Kill Lists. The document IDs to be killed can be specified in an auxiliary query defined by sql_query_killlist. The delta table must indicate the target tables for which the kill lists will be applied using the killlist_target directive. The impact of kill lists is permanent on the target table, meaning that even if a search is performed without the delta table, the suppressed documents will not appear in the search results.

Notice how we're overriding sql_query_pre in the delta source. We must explicitly include this override. If we don't, the REPLACE query would be executed during the delta source's build as well, effectively rendering it useless.

  • Example
# in MySQL
CREATE TABLE deltabreaker (
  index_name VARCHAR(50) NOT NULL,
  PRIMARY KEY (index_name)

# in manticore.conf
source main {
  sql_query_pre = REPLACE INTO deltabreaker SET index_name = 'main', created_at = NOW()
  sql_query =  SELECT id, title, UNIX_TIMESTAMP(updated_at) AS updated FROM documents WHERE deleted=0 AND  updated_at  >=FROM_UNIXTIME($start) AND updated_at  <=FROM_UNIXTIME($end)
  sql_query_range  = SELECT ( SELECT UNIX_TIMESTAMP(MIN(updated_at)) FROM documents) min, ( SELECT UNIX_TIMESTAMP(created_at)-1 FROM deltabreaker WHERE index_name='main') max
  sql_query_post_index = REPLACE INTO deltabreaker set index_name = 'delta', created_at = (SELECT created_at FROM deltabreaker t WHERE index_name='main')
  sql_attr_timestamp = updated

source delta : main {
  sql_query_pre =
  sql_query_range = SELECT ( SELECT UNIX_TIMESTAMP(created_at) FROM deltabreaker WHERE index_name='delta') min, UNIX_TIMESTAMP() max
  sql_query_killlist = SELECT id FROM documents WHERE updated_at >=  (SELECT created_at FROM deltabreaker WHERE index_name='delta')

table main {
  path = /var/lib/manticore/main
  source = main

table delta {
  path = /var/lib/manticore/delta
  source = delta
  killlist_target = main:kl

⪢ Adding data from tables

Merging tables

Merging two existing plain tables can be more efficient than indexing the data from scratch and might be desired in some cases (such as merging 'main' and 'delta' tables instead of simply rebuilding 'main' in the 'main+delta' partitioning scheme). Thus,indexer provides an option to do that. Merging tables is typically faster than rebuilding, but still not instant for huge tables. Essentially, it needs to read the contents of both tables once and write the result once. Merging a 100 GB and 1 GB table, for example, will result in 202 GB of I/O (but that's still likely less than indexing from scratch requires).

The basic command syntax is as follows:

sudo -u manticore indexer --merge DSTINDEX SRCINDEX [--rotate] [--drop-src]

Unless --drop-src is specified, only the DSTINDEX table will be affected: the contents of SRCINDEX will be merged into it.

The --rotate switch is required if DSTINDEX is already being served by searchd.

The typical usage pattern is to merge a smaller update from SRCINDEX into DSTINDEX. Thus, when merging attributes, the values from SRCINDEX will take precedence if duplicate document IDs are encountered. However, note that the "old" keywords will not be automatically removed in such cases. For example, if there's a keyword "old" associated with document 123 in DSTINDEX, and a keyword "new" associated with it in SRCINDEX, document 123 will be found by both keywords after the merge. You can supply an explicit condition to remove documents from DSTINDEX to mitigate this; the relevant switch is --merge-dst-range:

sudo -u manticore indexer --merge main delta --merge-dst-range deleted 0 0

This switch allows you to apply filters to the destination table along with merging. There can be several filters; all of their conditions must be met in order to include the document in the resulting merged table. In the example above, the filter passes only those records where 'deleted' is 0, eliminating all records that were flagged as deleted.

--drop-src enables dropping SRCINDEX after the merge and before rotating the tables, which is important if you specify DSTINDEX in killlist_target of DSTINDEX. Otherwise, when rotating the tables, the documents that have been merged into DSTINDEX may be suppressed by SRCINDEX.