Description:
When more than one block is used to store the result of a query in the query cache, the Querycache_stream object does not correctly use multiple blocks. The Querycache_stream::use_next_block() method never actually uses anything other than the first block.
How to repeat:
# We just want a small query cache, so we can fragment it easily
set GLOBAL query_cache_size=64*1024;
# This actually gives us a usable cache size of about 48K
# Each table is about 14K
create table t1 (a text);
insert into t1 values (repeat('abcdefghijklmnopqrstuvwxyz', 550));
create table t2 (a text);
insert into t2 values (repeat('ijklmnopqrstuvwxyzabcdefgh', 550));
# Load a query from each table into the query cache
--disable_result_log
select a from t1; # Q1
select a from t2; # Q2
--enable_result_log
show status like 'Qcache_%_blocks';
# Now the cache looks like (14K for Q1)(14K for Q2)(20K free)
# Flush Q1 from the cache by adding an out-of-order chunk to t1
insert into t1 select reverse(a) from t1;
show status like 'Qcache_%_blocks';
# Now the cache looks like (14K free)(14K for Q2)(20K free)
# Load our new data into the query cache
--disable_result_log
select a from t1; # Q3
--enable_result_log
show status like 'Qcache_%_blocks';
# Now the cache should be like (14K for Q3)(14K for Q2)(14K for Q3)(6K free)
# Note that Q3 is split across two chunks!
# Load Q3 from the cache, and actually pay attention to the results
select a from t1;
flush query cache;
drop table t1, t2;
Suggested fix:
Querycache_stream::use_next_block() needs to be fixed to actually make use of additional blocks and mark them appropriately. Querycache_stream::store_str_only() and Querycache_stream::load_str_only() also have off-by-one errors that cause them to call use_next_block() at the wrong time.
Description: When more than one block is used to store the result of a query in the query cache, the Querycache_stream object does not correctly use multiple blocks. The Querycache_stream::use_next_block() method never actually uses anything other than the first block. How to repeat: # We just want a small query cache, so we can fragment it easily set GLOBAL query_cache_size=64*1024; # This actually gives us a usable cache size of about 48K # Each table is about 14K create table t1 (a text); insert into t1 values (repeat('abcdefghijklmnopqrstuvwxyz', 550)); create table t2 (a text); insert into t2 values (repeat('ijklmnopqrstuvwxyzabcdefgh', 550)); # Load a query from each table into the query cache --disable_result_log select a from t1; # Q1 select a from t2; # Q2 --enable_result_log show status like 'Qcache_%_blocks'; # Now the cache looks like (14K for Q1)(14K for Q2)(20K free) # Flush Q1 from the cache by adding an out-of-order chunk to t1 insert into t1 select reverse(a) from t1; show status like 'Qcache_%_blocks'; # Now the cache looks like (14K free)(14K for Q2)(20K free) # Load our new data into the query cache --disable_result_log select a from t1; # Q3 --enable_result_log show status like 'Qcache_%_blocks'; # Now the cache should be like (14K for Q3)(14K for Q2)(14K for Q3)(6K free) # Note that Q3 is split across two chunks! # Load Q3 from the cache, and actually pay attention to the results select a from t1; flush query cache; drop table t1, t2; Suggested fix: Querycache_stream::use_next_block() needs to be fixed to actually make use of additional blocks and mark them appropriately. Querycache_stream::store_str_only() and Querycache_stream::load_str_only() also have off-by-one errors that cause them to call use_next_block() at the wrong time.