Thursday Dec 18, 2014

Long Class-Unloading Pauses with JDK8

Recently, I came across a report where a user was facing long GC pauses with JDK8. They didn't see this issue while running with JDK7. The problem with jdk8 was that sometimes the class unloading step of CMS-remark phase was taking a very long time to do its job. They were running with large pages enabled with both jdk7 and jdk8.

2014-11-30T01:51:35.632+0000: [GC (CMS Final Remark) [YG occupancy: 292006 K (1179648 K)][Rescan (parallel) , 0.0397261 secs][weak refs processing, 0.0954205 secs][class unloading, 29.2450253 secs][scrub symbol table, 4.9592268 secs][scrub string table, 0.0034572 secs][1 CMS-remark: 1101173K(1835008K)] 1393180K(3014656K), 34.3796360 secs] [Times: user=0.92 sys=0.38, real=34.37 secs]

2014-12-01T02:01:44.452+0000: [GC (CMS Final Remark) [YG occupancy: 550275 K (1179648 K)][Rescan (parallel) , 0.0785401 secs][weak refs processing, 0.0931797 secs][class unloading, 28.0163134 secs][scrub symbol table, 5.2806525 secs][scrub string table, 0.0034023 secs][1 CMS-remark: 1101982K(1835008K)] 1652257K(3014656K), 33.5087749 secs] [Times: user=1.48 sys=0.36, real=33.51 secs]

Now, here if we look closely, there is a huge difference in the times reported for 'user' 'sys' and 'real'. As we all know, 'user' is the amount of CPU time spent in user-mode within the process and 'sys' is the amount of CPU time spent in the kernel within the process. Other processes and the time the process spends blocked does not count towards these two figures. The 'real' time is the total elapsed time including time slices used by other processes and the time the process spends blocked (for example in waiting for I/O to complete, or system is performing memory activities such as paging/swapping).

In this case too, the problem seemed not be with the JVM or more specifically with the class unloading step but more with the GC threads getting blocked out. It turned out that the class unloading phase was causing lot of paging activity to occur at the system level which took away the CPU time from the GC threads.

What changed in the JVM from JDK7 to JDK8 is that in JDK8 we don't have PermGen. Instead we have MetaSpace which is allocated out of native memory space. As I mentioned earlier, the user was running with Large Pages enabled. With JDK7 by using option -XX:+UseLargePages, we get the large pages enabled for the Java Heap as well as for the PermGen. But with jdk8, with the class meta data stored in MetaSpace(native space), we don't get large pages enabled for the MetaSpace by default when we use +UseLargePages.

In this particular case, the system was configured to have higher number of large pages and a limited memory resource was left for the regular small pages. And a limited number of regular pages available for the MetaSpace caused paging activity during the class unloading phase leading to long GC pauses. The solution for this problem was to use -XX:+UseLargePagesInMetaspace that enables the large pages support for the MetaSpace. This avoided the paging activity and the class-unloading times returned to normal.

Please note that UseLargePagesInMetaspace is not enabled by default when UseLargePages is ON. This is intentional and is done to avoid any memory usage regressions. A classloader (and its classes metadata) is stored in a chunk of memory called Metachunk in the MetaSpace. To be able to hand back pages to the system, we have to clean out an entire page. It is possible that there may be many Metachunks of different classloaders present on a page, and that page can be returned to the system only when all the classloaders are dead and their Metachunks are not needed any more. This leads to delayed deallocation of pages and it is a bit more likely with large pages.

Another reason for not turning UseLargePagesInMetaspace option on by default is that we can not use large pages for the CompressedClass space. The problem there is that we can not dynamically grow the number of large pages; we have to pre-commit all the large pages when memory is reserved. CompressedClass space by default is 1GB of reserved memory and committing that 1GB up-front would cause memory usage regressions.

So, if one has the system configured in such a way that a larger part of memory is set aside for large pages and a relatively smaller part for the regular pages, it is likely to observe the above described long GC pauses during the clean up of MetaSpace. In such situation, using UseLargePagesInMetaspace may help resolve the problem.

Wednesday Jul 31, 2013

Troubleshooting Long GC Pauses

Low pause times during the application run is the most important goal for many enterprise applications, especially for the transaction-based systems where long latencies can result in the transaction time-outs. For systems running on the Java Virtual Machines, garbage collections can sometimes be the cause of the long pauses.

In this post I am going to describe different scenarios where we can encounter long GC pauses and how we can diagnose and troubleshoot these GC pauses.


Following are the different situations that can cause long GC pauses during the application run.

1. Fragmentation in the heap

Fragmentation in the Java Heap can cause GCs to occur more frequently and also sometimes causing long pauses in the GCs. This is more probable in the case of Concurrent Mark Sweep collector, also known as CMS, where the tenured generation space is not compacted with the concurrent collections.

In case of the CMS, due to fragmentation in the tenured generation space, the young generation collections can face promotion failures and thus triggering 'Concurrent Mode Failure' collections that are stop-the-world Full GCs, and Full GCs take a long time to finish as compared to the concurrent collection pauses.

Due to the fragmentation, the direct allocations in the tenured generation may fail even when there is lot of free space available and thus causing Full GCs. Fragmentation can also cause frequent allocation failures and thus triggering frequent Full GCs that increase the overall time the application is paused for.

The following logs collected with the CMS collector show that the fragmentation in the CMS generation space is very high, that leads to the promotion failure during a young generation ParNew collection and then a 'concurrent mode failure'. A Full GC is done in the event of 'concurrent mode failure' that takes a very long time, 17.1365396 seconds to finish.

{Heap before GC invocations=7430 (full 24):
par new generation total 134400K, used 121348K [0x53000000, 0x5c600000, 0x5c600000)
eden space 115200K, 99% used [0x53000000, 0x5a07e738, 0x5a080000)
from space 19200K, 32% used [0x5a080000, 0x5a682cc0, 0x5b340000)
to space 19200K, 0% used [0x5b340000, 0x5b340000, 0x5c600000)
concurrent mark-sweep generation total 2099200K, used 1694466K [0x5c600000, 0xdc800000, 0xdc800000)
concurrent-mark-sweep perm gen total 409600K, used 186942K [0xdc800000, 0xf5800000, 0xfbc00000)
10628.167: [GC Before GC:
Statistics for BinaryTreeDictionary:
Total Free Space: 103224160
Max Chunk Size: 5486
Number of Blocks: 57345
Av. Block Size: 1800
Tree Height: 36
Statistics for IndexedFreeLists:
Total Free Space: 371324
Max Chunk Size: 254
Number of Blocks: 8591 <---- High fragmentation
Av. Block Size: 43
free=103595484 frag=1.0000 <---- High fragmentation
Before GC:
Statistics for BinaryTreeDictionary:
Total Free Space: 0
Max Chunk Size: 0
Number of Blocks: 0
Tree Height: 0
Statistics for IndexedFreeLists:
Total Free Space: 0
Max Chunk Size: 0
Number of Blocks: 0
free=0 frag=0.0000
10628.168: [ParNew (promotion failed)
Desired survivor size 9830400 bytes, new threshold 1 (max 1)
- age 1: 4770440 bytes, 4770440 total
: 121348K->122157K(134400K), 0.4263254 secs]10628,594: [CMS10630.887: [CMS-concurrent-mark: 7.286/8.682 secs] [Times: user=14.81 sys=0.34, real=8.68 secs] (concurrent mode failure): 1698044K->625427K(2099200K), 17.1365396 secs] 1815815K->625427K(2233600K), [CMS Perm : 186942K->180711K(409600K)]After GC:
Statistics for BinaryTreeDictionary:
Total Free Space: 377269492
Max Chunk Size: 377269492
Number of Blocks: 1
Av. Block Size: 377269492
Tree Height: 1
Statistics for IndexedFreeLists:
Total Free Space: 0
Max Chunk Size: 0
Number of Blocks: 0
free=377269492 frag=0.0000
After GC:
Statistics for BinaryTreeDictionary:
Total Free Space: 0
Max Chunk Size: 0
Number of Blocks: 0
Tree Height: 0
Statistics for IndexedFreeLists:
Total Free Space: 0
Max Chunk Size: 0
Number of Blocks: 0
free=0 frag=0.0000
, 17.5645589 secs] [Times: user=17.82 sys=0.06, real=17.57 secs]
Heap after GC invocations=7431 (full 25):
par new generation total 134400K, used 0K [0x53000000, 0x5c600000, 0x5c600000)
eden space 115200K, 0% used [0x53000000, 0x53000000, 0x5a080000)
from space 19200K, 0% used [0x5b340000, 0x5b340000, 0x5c600000)
to space 19200K, 0% used [0x5a080000, 0x5a080000, 0x5b340000)
concurrent mark-sweep generation total 2099200K, used 625427K [0x5c600000, 0xdc800000, 0xdc800000)
concurrent-mark-sweep perm gen total 409600K, used 180711K [0xdc800000, 0xf5800000, 0xfbc00000)
Total time for which application threads were stopped: 17.5730653 seconds

2. Other OS activities happening at the time of GC

Sometimes the OS activities such as the swap space or networking activity happening at the time when GC is taking place can make the GC pauses last much longer. These pauses can be of the order of few seconds to some minutes.

If your system is configured to use swap space, Operating System may move inactive pages of memory of the JVM process to the swap space, to free up memory for the currently active process which may be the same process or a different process on the system. Swapping is very expensive as it requires disk accesses which are much slower as compared to the physical memory access. So, if during a garbage collection the system needs to perform swapping, the GC would seem to run for a very long time.

Following is the log of a young generation collection that lasts for 29.47 seconds.

{Heap before GC invocations=132 (full 0):
par new generation total 2696384K, used 2696384K [0xfffffffc20010000, 0xfffffffce0010000, 0xfffffffce0010000)
eden space 2247040K, 100% used [0xfffffffc20010000, 0xfffffffca9270000, 0xfffffffca9270000)
from space 449344K, 100% used [0xfffffffca9270000, 0xfffffffcc4940000, 0xfffffffcc4940000)
to space 449344K, 0% used [0xfffffffcc4940000, 0xfffffffcc4940000, 0xfffffffce0010000)
concurrent mark-sweep generation total 9437184K, used 1860619K [0xfffffffce0010000, 0xffffffff20010000, 0xffffffff20010000)
concurrent-mark-sweep perm gen total 1310720K, used 511451K [0xffffffff20010000, 0xffffffff70010000, 0xffffffff70010000)
2013-07-17T03:58:06.601-0700: 51522.120: [GC Before GC: : 2696384K->449344K(2696384K), 29.4779282 secs] 4557003K->2326821K(12133568K) ,29.4795222 secs] [Times: user=915.56 sys=6.35, real=29.48 secs]

Corresponding 'vmstat' output at 03:58:

kthr memory page disk faults cpu
r b w swap free re mf pi po fr de sr s0 s1 s2 s3 in sy cs us sy id
20130717_035806 0 0 0 77611960 94847600 55 266 0 0 0 0 0 0 0 0 0 3041 2644 2431 44 8 48
20130717_035815 0 0 0 76968296 94828816 79 324 0 18 18 0 0 0 0 1 0 3009 3642 2519 59 13 28
20130717_035831 1 0 0 77316456 94816000 389 2848 0 7 7 0 0 0 0 2 0 40062 78231 61451 42 6 53
20130717_035841 2 0 0 77577552 94798520 115 591 0 13 13 0 0 13 12 1 0 4991 8104 5413 2 0 98

This minor GC takes around 29 secs to complete. The corresponding vmstat output shows that the available swap space drops down by ~600mb during this period. That means during this garbage collection some pages from the RAM were moved out to the swap space, not necessarily by the same process running on the system.

From the above, it is clear that the physical memory available on the system is not enough for all the processes running on the system. The way to resolve this is to run fewer processes or if possible, add more RAM to increase the physical memory of the system. In the case above, the specified maximum tenured generation size is set as 9G and out of that only 1.8G is occupied. So it makes sense to reduce the heap size to lower the memory pressure on the physical memory so as to avoid or minimize the swapping activity.

Apart from swapping, we should monitor if there is any i/o or network activity happening during the long GC pauses. These can be monitored using iostat and netstat tools. It is also helpful to see the CPU statistics with the mpstat tool to figure out if enough CPU resources were available during the GC pauses.

3. Insufficient heap size

If the application footprint is larger than the maximum heap space that we have specified for the JVM, it results in frequent collections. Due to the insufficient heap space, the allocation requests fail and the JVM needs to invoke garbage collections in an attempt to reclaim space for the allocations. But since it cannot claim much space with each collection, subsequent allocation failures result in more GC invocations.

These frequent Full GCs cause long pauses in the application run. For example, in the following case, the permanent generation is almost full and the allocation attempts into the permanent generation are failing, triggering the Full GCs.

166687.013: [Full GC [PSYoungGen: 126501K->0K(922048K)] [PSOldGen: 2063794K->1598637K(2097152K)] 2190295K->1598637K(3019200K) [PSPermGen: 165840K->164249K(166016K)], 6.8204928 secs] [Times: user=6.80 sys=0.02, real=6.81 secs]
166699.015: [Full GC [PSYoungGen: 125518K->0K(922048K)] [PSOldGen: 1763798K->1583621K(2097152K)] 1889316K->1583621K(3019200K) [PSPermGen: 165868K->164849K(166016K)], 4.8204928 secs] [Times: user=4.80 sys=0.02, real=4.81 secs]

Similarly, the frequent Full GCs can occur if there is insufficient space in the tenured generation for the allocations or promotions.

The solution for these long pauses is to identify the average footprint of the application and then specify the heap size accordingly.

4. Bug in the JVM

Sometimes these long pauses could be due to a bug in the JVM. For example, due to the following bugs in the JVM, Java applications may face long GC pauses.

  • 6459113: CMS+ParNew: wildly different ParNew pause times depending on heap shape caused by allocation spread
    • fixed in JDK 6u1 and 7
  • 6572569: CMS: consistently skewed work distribution indicated in (long) re-mark pauses
    • fixed in JDK 6u4 and 7
  • 6631166: CMS: better heuristics when combatting fragmentation
    • fixed in JDK 6u21 and 7
  • 6999988: CMS: Increased fragmentation leading to promotion failure after CR#6631166 got implemented
    • fixed in JDK 6u25 and 7
  • 6683623: G1: use logarithmic BOT code such as used by other collectors
    • fixed in JDK 6u14 and 7
  • 6976350: G1: deal with fragmentation while copying objects during GC
    • fixed in JDK 8

If you are running with a JVM version affected with these bugs, please upgrade to the version where these bugs are fixed.

5. Explicit System GCs

Check if there are any explicit System GCs happening. Requests to invoke these System GCs which are stop-the-world Full GCs could be coming from the System.gc() calls from some class in the application or from a some third party module. These explicit System GCs too can cause very long pauses.

164638.058: [Full GC (System) [PSYoungGen: 22789K->0K(992448K)] [PSOldGen: 1645508K->1666990K(2097152K)] 1668298K->1666990K(3089600K) [PSPermGen: 164914K->164914K(166720K)], 5.7499132 secs] [Times: user=5.69 sys=0.06, real=5.75 secs]

If you are using RMI and are observing explicit Full GCs on a regular interval, then these are coming from the RMI implementation that triggers System.gc() on a regular interval. This interval can be configured using the following system properties:


The default value for these properties in JDK 1.4.2 and 5.0 is 60000 milliseconds, and 3600000 milliseconds in JDK 6 and later releases.

If you want to disable the explicit Full GCs invoked using System.gc(), run the application with -XX:+DisableExplicitGC JVM option.

How to approach the problem

1. Collect GC logs with -XX:+PrintGCDetails -XX:+PrintHeapAtGC -XX:+PrintGCTimeStamps -XX:+PrintGCDateStamps and -XX:+PrintGCApplicationStoppedTime. In case of the CMS collector, add option -XX:PrintFLSStatistics=2 as well.

The GC logs can give us details on the nature and the frequency of the GC pauses i.e. they can provide answers to the questions like - are the long GC pauses occurring during young collections or old collections, and how frequently those collections are encountering long pauses.

2. Monitor the overall health of the system using OS tools like vmstat, iostat, netstat and mpstat etc. on Solaris and Linux platforms, and tools like Process Monitor and Task Manager on the Windows operating system.

3. Use GCHisto tool to visually analyze the GC Logs and figure out which GCs are taking long time and if there is a pattern in the occurrence of these collections.

4. Try to see from the GC logs if there are any signs of fragmentation in the Java Heap space.

5. Monitor if the specified Heap size is enough to contain the footprint of the application.

6. Check if you are running with a JVM that has a known bug related to the long GC pauses and then upgrade if that bug is fixed in a later version.




« March 2015