Query.explain() method can provide useful insight when you are trying
to diagnose query performance issues and-or optimize queries. To examine how
your query is working, either embed the call inside your app (see:
Example 1), or use it interactively within a cblite shell (see:
The output from
Query.explain() remains the same whether invoked by an
app, or cblite — see Example 3 for an example of how it looks.
The Query Plan
The query plan section of the output displays a tabular form of the translated query's execution plan. It primarily shows how the data will be retrieved and, where appropriate, how it will be sorted for navigation and-or presentation purposes. For more on SQLite's Explain Query Plan — see: https://www.sqlite.org/eqp.html.
The query optimizer will attempt to retrieve the requested data items as efficiently as possible, which generally will be by using one or more of the available indexes. The retrieval method shows the approach decided upon by the optimizer — see Table 1.
|Search||Here the query is able to access the required data directly using keys into the index. Queries using the Search mode are the fastest.|
|Scan Index||Here the query is able to retrieve the data by scanning all or part-of the index (for example when seeking to match values within a range). This type of query is slower than search, but at least benefits from the compact and ordered form of the index.|
|Scan Table||Here the query must scan the database table(s) to retrieve the required data. It is the slowest of these methods and will benefit most from some form of optimization.|
When looking to optimize a query's retrieval method, consider whether:
- Providing an additional index makes sense.
- You could use an existing index — perhaps by restructuring the query to minimize wildcard use, or the reliance on functions that modify the query's interpretation of index keys (for example, 'lower').
- You could reduce the data set being requested to minimize the query's footprint on the database.
Order and Group
USE TEMP B-TREE FOR lines in the example indicate that the query requires
sorting to cater for grouping and then sorting again to present the output
results. Minimizing, if not eliminating, this ordering and re-ordering will
obviously reduce the amount of time taken to process your query.
Ask "is the grouping and-or ordering absolutely necessary?": if it isn't, drop it or modify it to minimize its impact.
Queries and Indexes
Before we begin querying documents, let's briefly mention the importance of having an appropriate and balanced approach to indexes.
Creating indexes can speed up the performance of queries. A query will typically return results more quickly if it can take advantage of an existing database index to search, narrowing down the set of documents to be examined.
Couchbase Lite for dart does not currently support partial value indexes; indexes with non-property expressions. You should only index with properties that you plan to use in the query.
The Query optimizer converts your query into a parse tree that groups zero or more and-connected clauses together (as dictated by your where conditionals) for effective query engine processing.
Ideally a query will be be able to satisfy its requirements entirely by either directly accessing the index or searching sequential index rows. Less good is if the query must scan the whole index; although the compact nature of most indexes means this is still much faster than the alternative of scanning the entire database with no help from the indexes at all.
Searches that begin with or rely upon an inequality with the primary key are inherently less effective than those using a primary key equality.
Working with the Query Optimizer
You may have noticed that sometimes a query runs faster on a second run, or after re-opening the database, or after deleting and recreating an index. This typically happens when SQL Query Optimizer has gathered sufficient stats to recognize a means of optimizing a sub-optimal query.
If only those stats were available from the start. In fact they are gathered after certain events, such as:
- Following index creation
- On a database close
- When running a database compact.
So, if your analysis of the Query Explain output indicates a sub-optimal query and your rewrites fail to sufficiently optimize it, consider compacting the database. Then re-generate the Query Explain and note any improvements in optimization. They may not, in themselves, resolve the issue entirely; but they can provide a uesful guide toward further optimizing changes you could make.
Wildcard and Like-based Queries
Like-based searches can use the index(es) only if:
- The search-string doesn't start with a wildcard.
- The primary search expression uses a property that is an indexed key.
- The search-string is a constant known at run time (that is, not a value derived during processing of the query).
To illustrate this we can use a modified query; replacing a simple equality test
In Example 5 we use a wildcard prefix and suffix. You can see that the
query plan decides on a retrieval method of
For more on indexes — see: Indexing.
By contrast, by removing the wildcard prefix
% (in Example 7), we see
that the query plan's retrieval method changes to become an index search. Where
practical, simple changes like this can make significant differences in query
Use Functions Wisely
Functions are a very useful tool in building queries, but be aware that they can impact whether the query-optimizer is able to use your index(es).
For example, you can observe a similar situation to that shown in
Wildcard and Like-based Queries when using
Function_.lower() on an indexed property.
Function_.lower(), changes things:
Knowing this, you can consider how you create the index; for example, using
Function_.lower() when you create the index and then always using
Try to minimize the amount of data retrieved. Reduce it down to the few properties you really do need to achieve the required result.
Consider fetching details lazily. You could break complex queries into components. Returning just the document IDs, then process the array of document IDs using either the Document API or a query thats uses the array of document IDs to return information.
Consider using paging to minimize the data returned when the number of results
returned is expected to be high. Getting the whole lot at once will be slow and
resource intensive: Plus does anyone want to access them all in one go? Instead
retrieve batches of information at a time, perhaps using
clauese to set a starting point for each subsequent batch.
Although, note that using query offsets becomes increasingly less effective as the overhead of skipping a growing number of rows each time increases. You can work around this, by instead using ranges of search-key values. If the last search-key value of batch one was 'x' then that could become the starting point for your next batch and-so-on.
Optimize document size in design. Smaller documents load more quickly. Break your data into logical linked units.
Consider Using Full Text Search instead of complex
REGEX patterns —
see Full Text Search.