Search timeouts
Although the Texpress search algorithm provides very high speed retrieval from large data sets, there are some classes of query that may take some time to execute.
In particular the following queries may execute slowly:
Search type |
Explanation |
---|---|
Search without any index support |
This class of searches includes querying on fields without index support, and wildcard searches where an index term has not been provided. For example searching for all surnames in the Parties module that start with |
Range searches |
While some index support is provided for range queries, it is only provided at the record descriptor level. A range search involves visiting every record descriptor and checking the range bits to see if the record falls within the search range. |
Large numbers of OR terms combined with AND terms |
When an OR query is executed the server logically performs one query per OR term, merging the results into one matching set. Since querying is very fast this is not a problem. However if a query contains a large number of OR terms and also some AND terms, the query optimi
is transformed by the optimi
If a large number of OR terms and AND terms exist, the optimi |
Since some of these queries may take some time on large data sets a facility was introduced with Texpress 8.0.028 that allows searches to be timed out: it is possible to stop a search after a given number of seconds. Support was also added to abort a search that would result in a large number of segments being retrieved (handles range queries and searches without index support).
The query timeout mechanism is supported in texforms, TexQL and texserver. Extensions have been added to texxmlserver to take advantage of the new facility.
Texpress options to control the timing out of searches include:
To control timing out of searches in Texpress, use:
timeout=x
where:
x
is the number of seconds a query will run before it is timed out
A value of zero will disable the timeout mechanism (which is the default value). After the required number of seconds has elapsed, a second option is consulted to determine how the timeout should be handled:
timeouthandler=[error|return|continue|ask]
where one of the values must be specified. The available handlers are:
|
return an error message indicating that the search has timed out. |
|
return any matching records (or none if none found) rather than an error. |
|
continue processing the search. A new timeout is not set. |
|
perform an upcall to get the user to respond (only available for texforms or texserver. TexQL same as error). |
The timeout mechanism is set at the application level; that is different timeouts cannot be set for individual tables. The reason for this is that a TexQL query may search over multiple tables and even virtual tables. If the option is set on a given table (e.g. by using dbnameopts), the value will override the application value for all searches. The default value for the handler is error.
If the handler is set to return,
any matches found are returned. Currently there is no mechanism to indicate whether the matches are a result of a full search or a timed out search.
In this example, a query is timed out after two seconds, returning an error in TexQL:
TEXPRESSOPTS="$TEXPRESSOPTS timeout=2 timeouthandler=error"
export TEXPRESSOPTS
texql
texql 1> select all from eparties;
Error: Your search has taken too long. Please adjust your query.
texql 2>
To add timeout settings for .profile-local
file in a client's directory. Pre ask
timeout handler, rather error
is generated.
In some instances, particularly web based searches, it would be useful to know that a search will result in a large number of matches before the search is performed. The searchlimit
option makes it possible to specify the maximum number of segments [a portion of the database] that may be searched by a given query. If the number of segments is large, it will take some time to process all the records to determine matches. Fortunately Texpress can determine very quickly whether a given query will result in a large number of segments being searched. Using this facility immediate feedback can be provided to users indicating a search is too vague.
The format of the option is:
searchlimit=[x|y%]
where:
x
is the maximum number of segments that can be searched. As an absolute number may not be useful (as it depends on the database configuration) a second form y%
is provided where y
is the maximum percentage of segments that may be searched.
Unlike the timeout
option, searchlimit
may be set on a table basis, so it is possible to restrict queries on a select number of tables, rather than just system-wide. For table specific settings use dbnameopts or the opts file in the table directory.
A handler can also be specified when the search limit is exceeded. The handler format is:
searchlimithandler=[error|return|continue|ask]
where the available settings are the same as for timeouthandler
. The default handler is error
.
In this example, the number of segments searched in eparties is limited to 50% and an error is returned in TexQL:
etaxonomyopts='searchlimit=50% searchlimithandler=error'
export etaxonomyopts
texql
texql 1> select all from eparties;
Error: Your search has too many matches. Please adjust your query.
texql 2>
To add search limit settings for .profile-local
file in a client?s directory. Pre ask
search limit handler, rather error
is generated.
Timeout extensions are also available for texxmlserver:
One of the reasons for introducing search timeouts was to restrict the resources made available for handling web based searches. Prior to introduction of the timeout facility a user could initiate a search that would take some time, tying up servers, even when the user had already given up on the search. The timeout mechanism provides three means for restricting access to resources via texxmlserver. Each of these is controlled by options set in the texxmlserver.conf file (for
Rather than setting timeout options in the .profile-local,
they can be set for texxmlserver only. Any setting in .profile-local
will apply to all instances of texforms, TexQL and texserver and so affect the Windows client. The texxmlserver.conf file allows the following options to be set:
Timeout=x
TimeoutHandler=[continue|error|return]
where:
x
is the number of seconds before a search is timed out. The default value is zero indicating there is no timeout activated.
The TimeoutHandler
option controls what happens when a timeout occurs. The values are the same as for the Texpress timeouthandler
setting. The timeout used affects all searches initiated by texxmlserver but does not affect update / insert / delete statements.
It is possible to restrict the number of segments that may be searched when querying. The restriction can be set for all tables or may be set on an individual table. The following settings may be used in the texxmlserver.conf
file:
SearchLimit=x[:dbname]
SearchLimit=y%[:dbname]
where:
x
|
is the maximum number of segments to search. |
y
|
is the maximum percentage of segments to search. |
If the setting is to apply to a table only, the value is appended with :dbname where dbname is the table name for which the restriction applies. More than one SearchLimit
may be specified. If more than one setting is specified, the order of the entries is significant. The System-wide entry should always be set first, followed by individual table settings. The default value is zero indicating there is no limit on the number of segments consulted for a search.
For example to only search 50% of segments for emultimedia, but 80% for eparties the following settings could be used:
SearchLimit=50%:emultimedia
SearchLimit=80%:eparties
Once the maximum number of segments is exceeded, an error is returned.
When performing a web based query it is possible to restrict the number of matching records. For example it may not make sense for a web based user to retrieve 20,000 records as they will never browse all the matches. The following setting may be used in the texxmlserver.conf file to restrict the number of matches returned:
MatchLimit=x[:dbname]
where:
x
is the maximum number of matches to return for a given query.
It is also possible to limit the number of matches on a table basis by specifying :dbname after the value, where dbname is the table name to restrict. More than one MatchLimit
setting may be specified. If more than one setting is specified, the order of the entries is significant. The System-wide entry should always be set first, followed by individual table settings. Once the maximum number of matches is reached the matching records are returned and the query terminated. The default value is -1
indicating there is no limit.