The backbone of the block, a modified LRU (least recently used) cache mapping SQL query strings to Statement objects.
Set the options for the statement cache. These are generally set at the database level using the :statement_cache_opts Database option.
:max_size | The maximum size (high water mark) for the cache. If an entry is added when the current size of the cache is equal to the maximum size, the cache is cleaned up to reduce the number of entries to the :min_size. Defaults to 1000. |
:min_size | The minimum size (low water mark) for the cache. On cleanup, the size of the cache is reduced to this number. Note that there could be fewer than this number of entries in the cache. Defaults to :max_size/2. |
:prepare_after | The number of executions to wait for before preparing |
the query server-side. If set to 1, prepares all executed queries server-side. If set to 5, does not attempt to prepare the query until the 5th execution. Defaults to 2.
:sorter | A callable object that takes two arguments, the current time and the related Statement instance, and should return some Comparable (usually a numeric) such that the lowest values returned are the first to be removed when it comes time to clean the pool. The default is basically: lambda{|t, stmt| (stmt.last_seen - t)/stmt.num_executes} so that it doesn’t remove statements that have been executed many times just because many less-frequently executed statements have been executed recently. |
The block passed is called with the Statement object’s name, only for statements that have been prepared, and should be used to deallocate the statements.
# File lib/sequel/extensions/pg_statement_cache.rb, line 108 108: def initialize(opts={}, &block) 109: @cleanup_proc = block 110: @prepare_after = opts.fetch(:prepare_after, 2) 111: @max_size = opts.fetch(:max_size, 1000) 112: @min_size = opts.fetch(:min_size, @max_size/2) 113: @sorter = opts.fetch(:sorter){method(:default_sorter)} 114: @ids = (1..@max_size).to_a.reverse 115: @hash = {} 116: # 117: # We add one so that when we clean the cache, the entry 118: # about to be added brings us to the min_size. 119: @size_diff = @max_size - @min_size + 1 120: end
Completely clear the statement cache, deallocating on the server side all statements that have been prepared.
# File lib/sequel/extensions/pg_statement_cache.rb, line 124 124: def clear 125: @hash.keys.each{|k| remove(k)} 126: end
Get the related statement name from the cache. If the entry is already in the cache, just bump it’s last seen time and the number of executions. Otherwise, add it to the cache. If the cache is already full, clean it up before adding it.
If the num of executions has passed the threshhold, yield the statement name to the block, which should be used to prepare the statement on the server side.
This method should return the prepared statment name if the statement has been prepared, and nil if the query has not been prepared and the statement should be executed normally.
# File lib/sequel/extensions/pg_statement_cache.rb, line 148 148: def fetch(sql) 149: unless stmt = @hash[sql] 150: # Get the next id from the id pool. 151: unless id = @ids.pop 152: # No id left, cache must be full, so cleanup and then 153: # get the next id from the id pool. 154: cleanup 155: id = @ids.pop 156: end 157: @hash[sql] = stmt = Statement.new(id) 158: end 159: 160: stmt.last_seen = Time.now 161: stmt.num_executes += 1 162: 163: if stmt.num_executes >= @prepare_after 164: if stmt.num_executes == @prepare_after 165: begin 166: yield(stmt.name) 167: rescue PGError 168: # An error occurred while preparing the statement, 169: # execute it normally (which will probably raise 170: # the error again elsewhere), but decrement the 171: # number of executions so we don't think we've 172: # prepared the statement when we haven't. 173: stmt.num_executes -= 1 174: return nil 175: end 176: end 177: stmt.name 178: end 179: end
After sorting the cache appropriately (so that the least important items are first), reduce the number of entries in the cache to the low water mark by removing the related query. Should only be called when the cache is full.
# File lib/sequel/extensions/pg_statement_cache.rb, line 201 201: def cleanup 202: t = Time.now 203: @hash.sort_by{|k,v| @sorter.call(t, v)}.first(@size_diff).each{|sql, stmt| remove(sql)} 204: end
Sort by time since last execution and number of executions. We don’t want to throw stuff out of the cache if it has been executed a lot, but a bunch of queries that were executed only once came in more recently.
# File lib/sequel/extensions/pg_statement_cache.rb, line 193 193: def default_sorter(t, stmt) 194: (stmt.last_seen - t)/stmt.num_executes 195: end
Remove the query from the cache. If it has been prepared, call the cleanup_proc to deallocate the statement.
# File lib/sequel/extensions/pg_statement_cache.rb, line 208 208: def remove(sql) 209: stmt = @hash.delete(sql) 210: if stmt.num_executes >= @prepare_after 211: @cleanup_proc.call(stmt.name) 212: end 213: 214: # Return id to the pool of ids 215: @ids.push(stmt.cache_id) 216: end
Disabled; run with --debug to generate this.
Generated with the Darkfish Rdoc Generator 1.1.6.