Memory leak in jdk1.7.0

Question

We've tried jdk 1.7.0_02. Ten days of running give 420 MB memory leak of the following objects:
- java.lang.management.MemoryUsage,
- [C (array of char),
- java.util.HashMap$Entry,
- [Ljava.util.HashMap$Entry (array of HashMap$Entry),
and some others.

This doesn't happen on jdk1.6.x.

First output of the "jmap -histo:live" command:

 num     #instances         #bytes  class name
----------------------------------------------
   1:        229527       14926888  [C
   2:        289290       13885920  java.lang.management.MemoryUsage
   3:        321029       10272928  java.util.HashMap$Entry
   4:         69923       10262184  <constMethodKlass>
   5:         69923        9527672  <methodKlass>
   6:          7048        7787040  <constantPoolKlass>
   7:        241693        7734176  java.lang.String
   8:          2038        5898408  [Ljava.util.concurrent.ConcurrentHashMap$HashEntry;
   9:          7048        5479056  <instanceKlassKlass>
  10:          5954        4499552  <constantPoolCacheKlass>
  11:         67844        4091672  [Ljava.util.HashMap$Entry;
  12:         41250        3942848  [B
  13:         65649        3151152  java.util.HashMap
  14:         71891        2875640  java.util.TreeMap$Entry
...
Total       2320965      138000120

The last output of the "jmap -histo:live" command done in 10 days after the first:

 num     #instances         #bytes  class name
----------------------------------------------
   1:       3147110      151061280  java.lang.management.MemoryUsage
   2:       3178875      101724000  java.util.HashMap$Entry
   3:       1087332       53822632  [C
   4:       1099503       35184096  java.lang.String
   5:        639442       31529224  [Ljava.util.HashMap$Entry;
   6:        637247       30587856  java.util.HashMap
   7:        629422       25176880  [Ljava.lang.management.MemoryUsage;
   8:        314711       17623816  com.sun.management.GcInfo
   9:         70107       10292776  <constMethodKlass>
  10:        631864       10109824  java.util.HashMap$EntrySet
  11:        314711       10070752  sun.management.GcInfoCompositeData
  12:         70107        9552696  <methodKlass>
  13:          7075        7817080  <constantPoolKlass>
  14:        314713        7554128  [Ljava.lang.Integer;
  15:          2048        5898744  [Ljava.util.concurrent.ConcurrentHashMap$HashEntry;
  16:          7075        5497200  <instanceKlassKlass>
  17:        315792        5052672  java.lang.Integer
  18:         47680        4912352  [B
...
Total      13206419      558217856

I also have 8 other histograms, made after each day of the tests. They show linear number of objects increasing. This is definitely not a noise. It is a stable leak 42 MB per day.

Did you observe similar behavior? In what scenarios? How did you cope with it?

Answer 1

Given the code in Java 7 is slightly different (but almost the same) as Java 6. I would expect to see these very subtle differences. I would take the snapshot after another 2 days to see if the JVM is still warming up. Connecting more monitoring clients would increase these values for example (as these are mostly monitoring objects).

If it is a memory leak, at 32K every two days, this could be wasting ~5 MB after a year of running and worth less than 5 cents of memory. Personally I would consider this too small to worry about.