I am using mmap:
INFO [main] 2011-09-29 21:35:21,458 DatabaseDescriptor.java (line
177) DiskAccessMode is mmap, indexAccessMode is mmap
I think the cleanerMethod must be OK now,
so the condition for GC
if (path == null
&& (DatabaseDescriptor.getDiskAccessMode() ==
Config.DiskAccessMode.mmap || DatabaseDescriptor.getIndexAccessMode()
== Config.DiskAccessMode.mmap)
&& !MmappedSegmentedFile.isCleanerAvailable())
{
has to be triggered by path==null,
I have put in more logging to debug it
On Thu, Sep 29, 2011 at 2:42 PM, Jonathan Ellis <jbel...@gmail.com> wrote:
> you're not using mmap i/o mode?
>
> On Thu, Sep 29, 2011 at 2:21 PM, Yang <teddyyyy...@gmail.com> wrote:
>> I see now :
>>
>> https://issues.apache.org/jira/browse/CASSANDRA-2521?focusedCommentId=13050396&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13050396
>>
>>
>>
>> but my logs doesn't show the error from
>>
>> MMappedSegmentedFile:
>>
>> public static void initCleaner()
>> {
>> try
>> {
>> cleanerMethod =
>> Class.forName("sun.nio.ch.DirectBuffer").getMethod("cleaner");
>> }
>> catch (Exception e)
>> {
>> // Perhaps a non-sun-derived JVM - contributions welcome
>> logger.info("Cannot initialize un-mmaper. (Are you using
>> a non-SUN JVM?) Compacted data files will not be removed promptly.
>> Consider using a SUN JVM or using standard disk access mode");
>> }
>> }
>>
>>
>> and indeed I'm using openJDK. I'll switch to sun jdk, but am still
>> curious why cleanerMethod is null in my case but I don't see the log
>> message
>>
>>
>> On Thu, Sep 29, 2011 at 12:03 PM, Yang <teddyyyy...@gmail.com> wrote:
>>> I thought the need for GC was based on the code that sstables
>>> deletions are tracked through weak/phantom references.
>>> now I see that sstable is deleted in SStableReader.ReleaseReferences()
>>> ---> SegmentedFile.cleanup(),
>>> if this is true, is GC still necessary?
>>>
>>> Thanks
>>> Yang
>>>
>>> On Thu, Sep 29, 2011 at 11:50 AM, Jonathan Ellis <jbel...@gmail.com> wrote:
>>>> As the comment says,
>>>> // retry after GCing to force unmap of compacted SSTables
>>>> so they can be deleted
>>>>
>>>> See also http://wiki.apache.org/cassandra/MemtableSSTable
>>>>
>>>> Note that this has changed in 1.0:
>>>> https://issues.apache.org/jira/browse/CASSANDRA-2521
>>>>
>>>> On Thu, Sep 29, 2011 at 1:14 PM, Yang <teddyyyy...@gmail.com> wrote:
>>>>> Table.getDataFileLocation() calls StorageService.requestGC(), which
>>>>> calls System.gc()
>>>>>
>>>>> I think I saw one GC in my log due to this code, and it caused a JVM
>>>>> freeze for 14 seconds.
>>>>> why do we need to call explicit GC ?
>>>>>
>>>>> thanks
>>>>> Yang
>>>>>
>>>>>
>>>>>
>>>>> [Times: user=2.81 sys=0.00, real=0.46 secs]
>>>>> 3472.943: [GC 3472.943: [ParNew: 21339139K->326984K(23592960K),
>>>>> 0.4742050 secs] 30764101K->10032251K(49807360K), 0.4743710 secs]
>>>>> [Times: user=2.91 sys=0.01, real=0.47 secs]
>>>>> 3519.043: [GC 3519.043: [ParNew: 21298504K->386099K(23592960K),
>>>>> 0.5201430 secs] 31003771K->10363088K(49807360K), 0.5203120 secs]
>>>>> [Times: user=3.15 sys=0.03, real=0.52 secs]
>>>>> 3560.277: [GC 3560.278: [ParNew: 21357619K->434850K(23592960K),
>>>>> 0.4693880 secs] 31334608K->10676370K(49807360K), 0.4695450 secs]
>>>>> [Times: user=2.92 sys=0.00, real=0.47 secs]
>>>>> 3601.343: [Full GC (System) 3601.343: [CMS:
>>>>> 10241519K->3496676K(26214400K), 14.6823750 secs]
>>>>> 28763882K->3496676K(49807360K), [CMS Perm : 27683K->27623K(27776K)],
>>>>> 14.6900120 secs] [Times: user=14.73 sys=0.00, real=14.69 secs]
>>>>> 3665.223: [GC 3665.224: [ParNew: 20971520K->361802K(23592960K),
>>>>> 0.2379040 secs] 24468196K->3858479K(49807360K), 0.2380660 secs]
>>>>> [Times: user=1.57 sys=0.01, real=0.23 secs]
>>>>> 3704.967: [GC 3704.968: [ParNew: 21333322K->359503K(23592960K),
>>>>> 0.4433910 secs] 24829999K->4173314K(49807360K), 0.4435500 secs]
>>>>> [Times: user=2.72 sys=0.02, real=0.44 secs]
>>>>> 3801.962: [GC 3801.962: [ParNew: 21331023K->388609K(23592960K),
>>>>> 0.5236180 secs] 25144834K->4479343K(49807360K), 0.5238240 secs]
>>>>> [Times: user=3.42 sys=0.00, real=0.53 secs]
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Jonathan Ellis
>>>> Project Chair, Apache Cassandra
>>>> co-founder of DataStax, the source for professional Cassandra support
>>>> http://www.datastax.com
>>>>
>>>
>>
>
>
>
> --
> Jonathan Ellis
> Project Chair, Apache Cassandra
> co-founder of DataStax, the source for professional Cassandra support
> http://www.datastax.com
>
