[CASSANDRA-13159] Coalescing strategy can enter infinite loop

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 2.1.17, 2.2.9, 3.0.11, 3.11.0
• Component/s: Legacy/Streaming and Messaging
• Severity: Normal
• Since Version: 2.1.0
• Assignee: Corentin Chary
• Reporter: Corentin Chary

Authors

Corentin Chary

Reviewers

Jeff Jirsa

• Created: 26/Jan/17 19:32
• Updated: 16/Apr/19 09:30
• Resolved: 10/Feb/17 23:33

Description

boolean maybeSleep(int messages, long averageGap, long maxCoalesceWindow, Parker parker)

maybeSleep() can enter an infinite loop if messages or averageGap ends up being 0 because sleep will be 0 and the while loop will never exit. I've noticed that on one of my clusters twice this week.

This can happen if in averageGap() sum is bigger than MEASURED_INTERVAL, which should be pretty rare but apparently happen to me.

Even if the diagnostic is wrong (and I'm pretty sure that this thread was using 100% CPU doing nothing), the fix seems pretty safe to apply.

diff --git a/src/java/org/apache/cassandra/utils/CoalescingStrategies.java b/src/java/org/apache/cassandra/utils/CoalescingStrategies.java
index 0aa980f..982d4a6 100644
--- a/src/java/org/apache/cassandra/utils/CoalescingStrategies.java
+++ b/src/java/org/apache/cassandra/utils/CoalescingStrategies.java
@@ -100,7 +100,7 @@ public class CoalescingStrategies
     {
         // only sleep if we can expect to double the number of messages we're sending in the time interval
         long sleep = messages * averageGap;
-        if (sleep > maxCoalesceWindow)
+        if (!sleep || sleep > maxCoalesceWindow)
             return false;
 
         // assume we receive as many messages as we expect; apply the same logic to the future batch:

Issue Links

• is duplicated by: CASSANDRA-13265 Expiration in OutboundTcpConnection can block the reader Thread