summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorMatthew Sackman <matthew@lshift.net>2009-08-28 12:48:56 +0100
committerMatthew Sackman <matthew@lshift.net>2009-08-28 12:48:56 +0100
commit85c7dd830e8a30e7b0b4c52889a896691e9b2fd1 (patch)
tree8f24288be205e847f267dafef838de7766c1fd0c
parent550f8691838add69c5dd6b8138f883377ce01d22 (diff)
downloadrabbitmq-server-git-85c7dd830e8a30e7b0b4c52889a896691e9b2fd1.tar.gz
cosmetics - documentation
-rw-r--r--src/rabbit_memory_manager.erl13
1 files changed, 8 insertions, 5 deletions
diff --git a/src/rabbit_memory_manager.erl b/src/rabbit_memory_manager.erl
index 91bcf1b878..aa3900e9f3 100644
--- a/src/rabbit_memory_manager.erl
+++ b/src/rabbit_memory_manager.erl
@@ -94,7 +94,7 @@
%% those processes and then going "whoops, didn't help after all"),
%% then we oppress the reporting process. When a process registers, it
%% can declare itself "unoppressable". If a process is unoppressable
-%% then it will not be sent to disk as a result of other processes
+%% then it will not be oppressed as a result of other processes
%% needing more tokens. However, if it itself needs additional tokens
%% which aren't available then it is still oppressed as before. This
%% feature is only used by the disk_queue, because if the disk queue
@@ -102,14 +102,17 @@
%% tight, the disk_queue would typically be one of the first processes
%% to be oppressed (sent to disk_only mode), which cripples
%% performance. Thus by setting it unoppressable, it is only possible
-%% for the disk_queue to be oppressed when it is active and
-%% attempting to increase its memory allocation.
+%% for the disk_queue to be oppressed when it is active and attempting
+%% to increase its memory allocation.
%%
%% If a process has been oppressed, it continues making memory
%% reports, as if it was liberated. As soon as a reported amount of
%% memory can be satisfied (and this can include oppressing other
-%% processes in the way described above), it will be liberated. We do
-%% not keep any information about oppressed processes.
+%% processes in the way described above), *and* the number of
+%% available tokens has changed by ?THRESHOLD_MULTIPLIER since the
+%% processes was oppressed, it will be liberated. This later condition
+%% prevents processes from continually oppressing each other if they
+%% themselves can be liberated by oppressing other processes.
%%
%% Note that the hibernate group can get very out of date. This is
%% fine, and somewhat unavoidable given the absence of useful APIs for