MapServer banner Home | Docs | Issue Tracker | FAQ | Download
en de es fr zh_cn it

MS RFC 65 - Single-pass Query Changes for 6.0

Date:2010/11/18
Author:Steve Lime
Contact:sdlime at comcast.net
Last Edited:2010-11-18
Status:Draft

Overview

This is a proposal to straighten out issues introduced in 5.6 with the single-pass query support. We probably introduced more change than necessary in hindsight. This RFC represents a simplification of both the C and MapScript APIs.

Single-pass query is really a misnomer. Pre-5.6 we would do an initial query (by point, bbox or whatever) to identify candidate shapes and then n small queries to retrieve each result. This was obviously very slow in some cases, particular with the RDBMS drivers. In 5.6 we switched things to basically hold a result set open post query and retain index values to the result set. We still make 2 passes through the results but this is much more efficient than before.

Initially we stuck the result set id in the shapeindex attribute of a result (of type resultCacheMemberObj). This index was later used to retrive the feature. Long story short, at the NYC sprint we discovered that some drivers were doing things differently. For Oracle, the result set ID was being passed in the tileindex and the shapeindex contained the global shape id. This was a reasonable thing to do and the PostGIS driver was changed at the sprint to behave the same way. So, the shapeindex contains the overall record index and the tileindex contains the index relative to the query result set. This re-use of the tileindex takes advantage of a an attribute that otherwise would go unused and saves us from further complicating things. Layers that use tile indexes are always file-based and don’t suffer (generally) from the performance problems what caused the changes in the first place.

Drivers that Implement msLayerResultsGetShape()

  • Oracle Spatial as msOracleSpatialLayerResultGetShape()
  • PostGIS as msPostGISLayerResultsGetShape()
  • OGR as msOGRLayerResultGetShape() (is identical to msOGRLayerGetShape())

Proposed Techincal Changes

Cleaning up the query result portion of the code.

  1. Drop the layer API function msLayerResultsGetShape() in favor of msLayerGetShape().
  2. Edit the msLayerGetShape() for drivers that support the result set indexes (now passed as a tile index value) to look for the tile index and if present work of the existing result set. If not, then work of the global id. That is, merge msLayerResultsGetShape() and msLayerGetShape() into a new msLayerGetShape().
  3. Remove resultsGetShape() and getFeature() methods from MapScript.
  4. Refactor getShape() MapScript method to a) return a shapeObj * like getFeature() does and to b) take a resultCacheMemberObj as input instead of the shape and tile indexes. That simplifies the inferface and abstracts the indexes so that we have more flexibility moving forward. We wouldn’t necessarily have to change the driver implementations to take a resultCacheMemberObj at this point, that could be done later.

The refactored getShape() signature would like so (in layer.i for SWIG):

%newobject getShape;
shapeObj *getShape(resultCacheMemberObj *r)

I’d also like to consider adding another query saving function to MapScript, something like saveQueryFeatures(). This function would write a pre-5.6 query file which consisted of a series of shape and tile indexes. We wouldn’t write tile indexes for layers that weren’t tiled (e.g. layers that didn’t have a TILEINDEX). This would restore some usefull funcitonallity that was removed with the 5.6 changes.

  1. Add signatures to the two styles of query files (query parameters and query results). This will allow us to write a single loader function and improve security.
  2. Add an optional flag to saveQuery() method in MapScript to save the actual query results (e.g. all the resultCacheMemberObj’s) instead of just the query parameters.

Post RFC a query action in MapScript would look like so:

...
my $rect = new mapscript::rectObj(420000, 5120000, 582000, 5200000);
$layer->queryByRect($map, $rect); # layer is left open after this operation

for(my $i=0; $i<$layer->getNumResults(); $i++) {
       my $shape = $layer->getShape($layer->getResult($)); # much simpler
       ...
}

# save query
$map->saveQuery('myquery.qy'); # new style (query parameters)
$map->saveQuery('myquery.qy', 1); # old style (query feature indexes)

The changes are relatively technically mild. The changes in MapScript are big in that scripts that process query results will need an update. In the end I think that’s worth it and this will be more intuitive for users.

Bug ID

#3647

Voting history

Passed with a +1 from Steve L., Tom, Jeff, Steve W., Perry.