MS RFC 109: Optimizing Runtime Substitutions¶
- Last Updated
While profiling the WMS shootout queries, a noticeable amount of time was spent in the Run-time Substitution code, even though the functionality was not being used at all. As runtime substitutions apply to all mapserv requests, an optimization in this part of the code is beneficial for all mapserv operations.
For some specific cases, there can be the need to apply runtime substitutions inside metadata entries (e.g. modify ows_enable_request when using a proxy between mapserver and the client, to enable specific operations for authenticated clients). The current implementation prohibits this as it would put a far too noticeable hit on performance.
2. Proposed Rewrite¶
In pseudo-code, the current algorithm is: (KVP stands for key-value-pair, and is one of the parameters passed in the URL to mapserver, e.g. …&map=path/to/mapfile.map&…)
foreach KVP: foreach outputformat: if KVP appears in "FILENAME" formatoption: if KVP validates against MAP_>WEB->VALIDATION: apply substitution foreach layer: foreach class: if KVP appears in a substitutable CLASS keyword (EXPRESSION, TEXT, TITLE): if KVP validates against CLASS->VALIDATION: apply substitution else if KVP validates against LAYER->VALIDATION: apply substitution else if KVP validates against MAP->WEB->VALIDATION: apply substitution if KVP appears in a substitutable LAYER keyword (DATA, CONNECTION, FILTER, TILEINDEX): else if KVP validates against LAYER->VALIDATION: apply substitution else if KVP validates against MAP->WEB->VALIDATION: apply substitution
For a typical request where the number of KVPs is relatively high (e.g. all OWS requests) and where the number of layers/classes is also relatively high, this results in a very large number of strcasestr calls being made for the “if KVP appears in XXX keyword” parts. These happen whether or not actual runtime substitutions are going to be used or not.
With this RFC, we will be switching to this pseudo-code:
foreach layer: foreach class: foreach class->validation: if validation_key in kvp: if kvp validates: apply class substitutions foreach layer->validation: if validation_key in kvp: if kvp validates: apply layer substitutions foreach class: apply class substitutions foreach map->web->validation: if validation_key in kvp: if kvp validates: apply outputformat validations foreach layer: apply layer substitutions foreach class: apply class substitutions
This has the advantage of requiring very little overhead if no substitutions are in use, and somewhat simplifying the code. This optimization becomes possible because we require VALIDATION entries to be present (since 6.0).
In exchange for this speedup, we now allow substitutions to occur on layer->metadata and map->web->metadata entries, which would have had a too large impact on performance with the previous method. We can also imagine adding substitutions to other mapfile elements, now that the cost of this operation does not impact performance when not in use.
For performance reasons, validations should be put in the validation block “closest” to where they will actually be used. Putting a validation in map->web->validation is practical as it will apply to the whole mapfile, it however will behave slightly slower than duplicating it in the needed layer/class validations.
Update from 2015/02, with +1 from ThomasB, StephanM, DanielM, MikeS and SteveW: As discussed on the email list, this RFC is extended with the two following features:
Allow substitutions to happen in PROCESSING entries (needed for RFC91 native filters, could also be used to e.g. allow to dynamically set resampling methods for raster layers)
Allow substitution values to be read in priority from an envirronment variable. This would allow a server administrator to set substitutions based on a user id, etc… As an envirronment lookup would only happen if the mapfile contains the corresponding entry in a VALIDATION block, this feature would not allow the global environment to be leaked, unless the user has chosen a substitution key that matches an existing environment value. Note that a key from the envirronment would take precedence over the same key passed in the URL.
3.1 Backwards Incompatibilities¶
As substitutions can now apply to metadata, there is a potential incompatibility for this uncommon scenario:
a metadata block contains a %variable% substring in one of its entries
and the mapfile contains a VALIDATION block for the “variable” runsub.
In this case, a substitution will occur if variable=foo is present in the URL, whereas this would not have been the case beforehand.
3.2 Issue Tracking ID¶
3.3 Voting History¶
+1 from ThomasB, TomK, SteveW, YewondwossenA, SteveL, PerryN, MikeS, JeffM and StephanM