You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@lucene.apache.org by "Johannes Brucher (Jira)" <ji...@apache.org> on 2020/02/10 11:11:00 UTC
[jira] [Updated] (SOLR-14251) Shard Split on HDFS
[ https://issues.apache.org/jira/browse/SOLR-14251?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Johannes Brucher updated SOLR-14251:
------------------------------------
Description:
Shard Split on HDFS Index will evaluate local disc space instead of HDFS space
When performing a shard split on an index that is stored on HDFS the SplitShardCmd however evaluates the free disc space on the local file system of the server where Solr is installed.
SplitShardCmd assumes that its main phase (when the Lucene index is being split) always executes on the local file system of the shard leader; and indeed the ShardSplitCmd.checkDiskSpace() checks the local file system's free disk space - even though the actual data is written to the HDFS Directory so it (almost) doesn't affect the local FS (except for core.properties file).
See also: [https://lucene.472066.n3.nabble.com/HDFS-Shard-Split-td4449920.html]
My setup to reproduce the issue:
* Solr deployed on Openshift with local disc of about 5GB
* HDFS configuration based on solrconfig.xml with
{code:java}
<directoryFactory name="DirectoryFactory" class="solr.HdfsDirectoryFactory">
<str name="solr.hdfs.home">hdfs://path/to/index/</str>
...
{code}
* Split command:
{code:java}
.../admin/collections?action=SPLITSHARD&collection=collection1&shard=shard1&async=1234{code}
* Response:
{code:java}
{
"responseHeader":{"status":0,"QTime":32},
"Operation splitshard caused exception:":"org.apache.solr.common.SolrException:org.apache.solr.common.SolrException: not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
"exception":{
"msg":"not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
"rspCode":500},
"status":{"state":"failed","msg":"found [1234] in failed tasks"}
}
{code}
was:
Shard Split on HDFS Index will evaluate local disc space instead of HDFS space
When performing a shard split on an index that is stored on HDFS the SplitShardCmd however evaluates the free disc space on the local file system of the server where Solr is installed.
SplitShardCmd assumes that its main phase (when the Lucene index is being split) always executes on the local file system of the shard leader; and indeed the ShardSplitCmd.checkDiskSpace() checks the local file system's free disk space - even though the actual data is written to the HDFS Directory so it (almost) doesn't affect the local FS (except for core.properties file).
See also: [https://lucene.472066.n3.nabble.com/HDFS-Shard-Split-td4449920.html]
My setup to reproduce the issue:
* Solr deployed on Openshift with local disc of about 5GB
* HDFS configuration based on solrconfig.xml with
{code:java}
<directoryFactory name="DirectoryFactory" class="solr.HdfsDirectoryFactory">
<str name="solr.hdfs.home">hdfs://path/to/index/str>
...
{code}
* Split command:
{code:java}
.../admin/collections?action=SPLITSHARD&collection=collection1&shard=shard1&async=1234{code}
* Response:
{code:java}
{
"responseHeader":{"status":0,"QTime":32},
"Operation splitshard caused exception:":"org.apache.solr.common.SolrException:org.apache.solr.common.SolrException: not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
"exception":{
"msg":"not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
"rspCode":500},
"status":{"state":"failed","msg":"found [1234] in failed tasks"}
}
{code}
> Shard Split on HDFS
> --------------------
>
> Key: SOLR-14251
> URL: https://issues.apache.org/jira/browse/SOLR-14251
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Components: hdfs
> Affects Versions: 8.4
> Reporter: Johannes Brucher
> Priority: Major
>
> Shard Split on HDFS Index will evaluate local disc space instead of HDFS space
> When performing a shard split on an index that is stored on HDFS the SplitShardCmd however evaluates the free disc space on the local file system of the server where Solr is installed.
> SplitShardCmd assumes that its main phase (when the Lucene index is being split) always executes on the local file system of the shard leader; and indeed the ShardSplitCmd.checkDiskSpace() checks the local file system's free disk space - even though the actual data is written to the HDFS Directory so it (almost) doesn't affect the local FS (except for core.properties file).
> See also: [https://lucene.472066.n3.nabble.com/HDFS-Shard-Split-td4449920.html]
> My setup to reproduce the issue:
> * Solr deployed on Openshift with local disc of about 5GB
> * HDFS configuration based on solrconfig.xml with
> {code:java}
> <directoryFactory name="DirectoryFactory" class="solr.HdfsDirectoryFactory">
> <str name="solr.hdfs.home">hdfs://path/to/index/</str>
> ...
> {code}
> * Split command:
> {code:java}
> .../admin/collections?action=SPLITSHARD&collection=collection1&shard=shard1&async=1234{code}
> * Response:
> {code:java}
> {
> "responseHeader":{"status":0,"QTime":32},
> "Operation splitshard caused exception:":"org.apache.solr.common.SolrException:org.apache.solr.common.SolrException: not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
> "exception":{
> "msg":"not enough free disk space to perform index split on node <solr instance>:8983_solr, required: 294.64909074269235, available: 5.4632568359375",
> "rspCode":500},
> "status":{"state":"failed","msg":"found [1234] in failed tasks"}
> }
> {code}
>
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org