Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
Hey Shawn / Solr User Group, This makes perfect sense to me. Thanks for the thorough answer. The CSV update handler works at a lower level than the DataImport handler, and doesn't have clean or full-import options, which defaults to clean=true. The DIH is like a full application embedded inside Solr, one that uses an update handler -- it is not itself an update handler. When clean=true or using full-import without a clean option, DIH itself sends a delete all documents update request. And similiarly, my assumption is in the event of a non-syntactical failure/interuption (such as a server crash) during the CSV Update a rollback (stream.body=rollback/) would also need to be manually requested (or automatted but outside of Solr) where as the DIH automates this Request on my behalf as well...? Is there anyway to detect this failure or interuption?...A real example is, I was in the process of indexing data via the CSV Update and somebody bounced the server before it completed. No actual errors were produced but it appeared that the CSV Update process stopped at the point of the reboot. My assumption is, if I had passed in a rollback, I'd get the previously indexed data , given I didn't request a delete beforehand (haven't yet tested this). But wondering, how I could automatically detect this? This I guess is where DIH starts gaining some merit. Also - the response that the DIH produces when the indexing process is complete appears to be a lot more mature in that it explicity suggest the index completed and that information can can be re-queried. It would be nice if the CSV Update provided a similiar response..my assumption is it would first need to know how many lines exist on the file in order to know whether or not the job actually completed... Also - outside of solr initiating a delete due to encountering the same UniqueKey, is there anything else that could cause a delete to be initiated by Solr? Lastly, is there any concern of running multiple Update CSV requests on different data files containing different data? Thanks in advance. This was very helpful. Mike From: Shawn Heisey s...@elyograg.org To: solr-user@lucene.apache.org Sent: Monday, July 1, 2013 2:30 PM Subject: Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 On 7/1/2013 12:56 PM, Mike L. wrote: Hey Ahmet / Solr User Group, I tried using the built in UpdateCSV and it runs A LOT faster than a FileDataSource DIH as illustrated below. However, I am a bit confused about the numDocs/maxDoc values when doing an import this way. Here's my Get command against a Tab delimted file: (I removed server info and additional fields.. everything else is the same) http://server:port/appname/solrcore/update/csv?commit=trueheader=falseseparator=%09escape=\stream.file=/location/of/file/on/server/file.csvfieldnames=id,otherfields My response from solr ?xml version=1.0 encoding=UTF-8? response lst name=responseHeaderint name=status0/intint name=QTime591/int/lst /response I am experimenting with 2 csv files (1 with 10 records, the other with 1000) to see If I can get this to run correctly before running my entire collection of data. I initially loaded the first 1000 records to an empty core and that seemed to work, however, but when running the above with a csv file that has 10 records, I would like to see only 10 active records in my core. What I get instead, when looking at my stats page: numDocs 1000 maxDoc 1010 If I run the same url above while appending an 'optimize=true', I get: numDocs 1000, maxDoc 1000. A discrepancy between numDocs and maxDoc indicates that there are deleted documents in your index. You might already know this, so here's an answer to what I think might be your actual question: If you want to delete the 1000 existing documents before adding the 10 documents, then you have to actually do that deletion. The CSV update handler works at a lower level than the DataImport handler, and doesn't have clean or full-import options, which defaults to clean=true. The DIH is like a full application embedded inside Solr, one that uses an update handler -- it is not itself an update handler. When clean=true or using full-import without a clean option, DIH itself sends a delete all documents update request. If you didn't already know the bit about the deleted documents, then read this: It can be normal for indexing new documents to cause deleted documents. This happens when you have the same value in your UniqueKey field as documents that are already in your index. Solr knows by the config you gave it that they are the same document, so it deletes the old one before adding the new one. Solr has no way to know whether the document it already had or the document you are adding is more current, so it assumes you know what you are doing and takes care of the deletion for you. When you optimize your index
Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
The split/group implementation in RegexTransformer is not as efficient as CSVLoader. Perhaps we need a specialized csv loader in DIH. SOLR-2549 aims to add this support. I'll take a look. On Tue, Jul 2, 2013 at 12:26 AM, Mike L. javaone...@yahoo.com wrote: Hey Ahmet / Solr User Group, I tried using the built in UpdateCSV and it runs A LOT faster than a FileDataSource DIH as illustrated below. However, I am a bit confused about the numDocs/maxDoc values when doing an import this way. Here's my Get command against a Tab delimted file: (I removed server info and additional fields.. everything else is the same) http://server:port/appname/solrcore/update/csv?commit=trueheader=falseseparator=%09escape=\stream.file=/location/of/file/on/server/file.csvfieldnames=id,otherfields My response from solr ?xml version=1.0 encoding=UTF-8? response lst name=responseHeaderint name=status0/intint name=QTime591/int/lst /response I am experimenting with 2 csv files (1 with 10 records, the other with 1000) to see If I can get this to run correctly before running my entire collection of data. I initially loaded the first 1000 records to an empty core and that seemed to work, however, but when running the above with a csv file that has 10 records, I would like to see only 10 active records in my core. What I get instead, when looking at my stats page: numDocs 1000 maxDoc 1010 If I run the same url above while appending an 'optimize=true', I get: numDocs 1000, maxDoc 1000. Perhaps the commit=true is not doing what its supposed to or am I missing something? I also trying passing a commit afterward like this: http://server:port/appname/solrcore/update?stream.body=%3Ccommit/%3E ( didn't seem to do anything either) From: Ahmet Arslan iori...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org; Mike L. javaone...@yahoo.com Sent: Saturday, June 29, 2013 7:20 AM Subject: Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 Hi Mike, You could try http://wiki.apache.org/solr/UpdateCSV And make sure you commit at the very end. From: Mike L. javaone...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org Sent: Saturday, June 29, 2013 3:15 AM Subject: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 I've been working on improving index time with a JdbcDataSource DIH based config and found it not to be as performant as I'd hoped for, for various reasons, not specifically due to solr. With that said, I decided to switch gears a bit and test out FileDataSource setup... I assumed by eliminiating network latency, I should see drastic improvements in terms of import time..but I'm a bit surprised that this process seems to run much slower, at least the way I've initially coded it. (below) The below is a barebone file import that I wrote which consumes a tab delimited file. Nothing fancy here. The regex just seperates out the fields... Is there faster approach to doing this? If so, what is it? Also, what is the recommended approach in terms of index/importing data? I know thats may come across as a vague question as there are various options available, but which one would be considered the standard approach within a production enterprise environment. (below has been cleansed) dataConfig dataSource name=file type=FileDataSource / document entity name=entity1 processor=LineEntityProcessor url=[location_of_file]/file.csv dataSource=file transformer=RegexTransformer,TemplateTransformer field column=rawLine regex=^(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)$ groupNames=field1,field2,field3,field4,field5,field6,field7,field8,field9,field10,field11,field12,field13,field14,field15,field16,field17,field18,field19,field10,field11,field12 / /entity /document /dataConfig Thanks in advance, Mike Thanks in advance, Mike -- Regards, Shalin Shekhar Mangar.
Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
Hey Ahmet / Solr User Group, I tried using the built in UpdateCSV and it runs A LOT faster than a FileDataSource DIH as illustrated below. However, I am a bit confused about the numDocs/maxDoc values when doing an import this way. Here's my Get command against a Tab delimted file: (I removed server info and additional fields.. everything else is the same) http://server:port/appname/solrcore/update/csv?commit=trueheader=falseseparator=%09escape=\stream.file=/location/of/file/on/server/file.csvfieldnames=id,otherfields My response from solr ?xml version=1.0 encoding=UTF-8? response lst name=responseHeaderint name=status0/intint name=QTime591/int/lst /response I am experimenting with 2 csv files (1 with 10 records, the other with 1000) to see If I can get this to run correctly before running my entire collection of data. I initially loaded the first 1000 records to an empty core and that seemed to work, however, but when running the above with a csv file that has 10 records, I would like to see only 10 active records in my core. What I get instead, when looking at my stats page: numDocs 1000 maxDoc 1010 If I run the same url above while appending an 'optimize=true', I get: numDocs 1000, maxDoc 1000. Perhaps the commit=true is not doing what its supposed to or am I missing something? I also trying passing a commit afterward like this: http://server:port/appname/solrcore/update?stream.body=%3Ccommit/%3E ( didn't seem to do anything either) From: Ahmet Arslan iori...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org; Mike L. javaone...@yahoo.com Sent: Saturday, June 29, 2013 7:20 AM Subject: Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 Hi Mike, You could try http://wiki.apache.org/solr/UpdateCSV And make sure you commit at the very end. From: Mike L. javaone...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org Sent: Saturday, June 29, 2013 3:15 AM Subject: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 I've been working on improving index time with a JdbcDataSource DIH based config and found it not to be as performant as I'd hoped for, for various reasons, not specifically due to solr. With that said, I decided to switch gears a bit and test out FileDataSource setup... I assumed by eliminiating network latency, I should see drastic improvements in terms of import time..but I'm a bit surprised that this process seems to run much slower, at least the way I've initially coded it. (below) The below is a barebone file import that I wrote which consumes a tab delimited file. Nothing fancy here. The regex just seperates out the fields... Is there faster approach to doing this? If so, what is it? Also, what is the recommended approach in terms of index/importing data? I know thats may come across as a vague question as there are various options available, but which one would be considered the standard approach within a production enterprise environment. (below has been cleansed) dataConfig dataSource name=file type=FileDataSource / document entity name=entity1 processor=LineEntityProcessor url=[location_of_file]/file.csv dataSource=file transformer=RegexTransformer,TemplateTransformer field column=rawLine regex=^(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)$ groupNames=field1,field2,field3,field4,field5,field6,field7,field8,field9,field10,field11,field12,field13,field14,field15,field16,field17,field18,field19,field10,field11,field12 / /entity /document /dataConfig Thanks in advance, Mike Thanks in advance, Mike
Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
On 7/1/2013 12:56 PM, Mike L. wrote: Hey Ahmet / Solr User Group, I tried using the built in UpdateCSV and it runs A LOT faster than a FileDataSource DIH as illustrated below. However, I am a bit confused about the numDocs/maxDoc values when doing an import this way. Here's my Get command against a Tab delimted file: (I removed server info and additional fields.. everything else is the same) http://server:port/appname/solrcore/update/csv?commit=trueheader=falseseparator=%09escape=\stream.file=/location/of/file/on/server/file.csvfieldnames=id,otherfields My response from solr ?xml version=1.0 encoding=UTF-8? response lst name=responseHeaderint name=status0/intint name=QTime591/int/lst /response I am experimenting with 2 csv files (1 with 10 records, the other with 1000) to see If I can get this to run correctly before running my entire collection of data. I initially loaded the first 1000 records to an empty core and that seemed to work, however, but when running the above with a csv file that has 10 records, I would like to see only 10 active records in my core. What I get instead, when looking at my stats page: numDocs 1000 maxDoc 1010 If I run the same url above while appending an 'optimize=true', I get: numDocs 1000, maxDoc 1000. A discrepancy between numDocs and maxDoc indicates that there are deleted documents in your index. You might already know this, so here's an answer to what I think might be your actual question: If you want to delete the 1000 existing documents before adding the 10 documents, then you have to actually do that deletion. The CSV update handler works at a lower level than the DataImport handler, and doesn't have clean or full-import options, which defaults to clean=true. The DIH is like a full application embedded inside Solr, one that uses an update handler -- it is not itself an update handler. When clean=true or using full-import without a clean option, DIH itself sends a delete all documents update request. If you didn't already know the bit about the deleted documents, then read this: It can be normal for indexing new documents to cause deleted documents. This happens when you have the same value in your UniqueKey field as documents that are already in your index. Solr knows by the config you gave it that they are the same document, so it deletes the old one before adding the new one. Solr has no way to know whether the document it already had or the document you are adding is more current, so it assumes you know what you are doing and takes care of the deletion for you. When you optimize your index, deleted documents are purged, which is why the numbers match there. Thanks, Shawn
Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
Hi Mike, You could try http://wiki.apache.org/solr/UpdateCSV And make sure you commit at the very end. From: Mike L. javaone...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org Sent: Saturday, June 29, 2013 3:15 AM Subject: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 I've been working on improving index time with a JdbcDataSource DIH based config and found it not to be as performant as I'd hoped for, for various reasons, not specifically due to solr. With that said, I decided to switch gears a bit and test out FileDataSource setup... I assumed by eliminiating network latency, I should see drastic improvements in terms of import time..but I'm a bit surprised that this process seems to run much slower, at least the way I've initially coded it. (below) The below is a barebone file import that I wrote which consumes a tab delimited file. Nothing fancy here. The regex just seperates out the fields... Is there faster approach to doing this? If so, what is it? Also, what is the recommended approach in terms of index/importing data? I know thats may come across as a vague question as there are various options available, but which one would be considered the standard approach within a production enterprise environment. (below has been cleansed) dataConfig dataSource name=file type=FileDataSource / document entity name=entity1 processor=LineEntityProcessor url=[location_of_file]/file.csv dataSource=file transformer=RegexTransformer,TemplateTransformer field column=rawLine regex=^(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)$ groupNames=field1,field2,field3,field4,field5,field6,field7,field8,field9,field10,field11,field12,field13,field14,field15,field16,field17,field18,field19,field10,field11,field12 / /entity /document /dataConfig Thanks in advance, Mike
Re: FileDataSource vs JdbcDataSouce (speed) Solr 3.5
Mike: One issue is that you're forcing all the work onto the Solr server, and single-threading to boot by using DIH. You can consider moving to a SolrJ model where you can have N clients sending data to Solr if you can partition the data up amongst the N clients cleanly. FWIW, Erick On Sat, Jun 29, 2013 at 8:20 AM, Ahmet Arslan iori...@yahoo.com wrote: Hi Mike, You could try http://wiki.apache.org/solr/UpdateCSV And make sure you commit at the very end. From: Mike L. javaone...@yahoo.com To: solr-user@lucene.apache.org solr-user@lucene.apache.org Sent: Saturday, June 29, 2013 3:15 AM Subject: FileDataSource vs JdbcDataSouce (speed) Solr 3.5 I've been working on improving index time with a JdbcDataSource DIH based config and found it not to be as performant as I'd hoped for, for various reasons, not specifically due to solr. With that said, I decided to switch gears a bit and test out FileDataSource setup... I assumed by eliminiating network latency, I should see drastic improvements in terms of import time..but I'm a bit surprised that this process seems to run much slower, at least the way I've initially coded it. (below) The below is a barebone file import that I wrote which consumes a tab delimited file. Nothing fancy here. The regex just seperates out the fields... Is there faster approach to doing this? If so, what is it? Also, what is the recommended approach in terms of index/importing data? I know thats may come across as a vague question as there are various options available, but which one would be considered the standard approach within a production enterprise environment. (below has been cleansed) dataConfig dataSource name=file type=FileDataSource / document entity name=entity1 processor=LineEntityProcessor url=[location_of_file]/file.csv dataSource=file transformer=RegexTransformer,TemplateTransformer field column=rawLine regex=^(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)$ groupNames=field1,field2,field3,field4,field5,field6,field7,field8,field9,field10,field11,field12,field13,field14,field15,field16,field17,field18,field19,field10,field11,field12 / /entity /document /dataConfig Thanks in advance, Mike
FileDataSource vs JdbcDataSouce (speed) Solr 3.5
I've been working on improving index time with a JdbcDataSource DIH based config and found it not to be as performant as I'd hoped for, for various reasons, not specifically due to solr. With that said, I decided to switch gears a bit and test out FileDataSource setup... I assumed by eliminiating network latency, I should see drastic improvements in terms of import time..but I'm a bit surprised that this process seems to run much slower, at least the way I've initially coded it. (below) The below is a barebone file import that I wrote which consumes a tab delimited file. Nothing fancy here. The regex just seperates out the fields... Is there faster approach to doing this? If so, what is it? Also, what is the recommended approach in terms of index/importing data? I know thats may come across as a vague question as there are various options available, but which one would be considered the standard approach within a production enterprise environment. (below has been cleansed) dataConfig dataSource name=file type=FileDataSource / document entity name=entity1 processor=LineEntityProcessor url=[location_of_file]/file.csv dataSource=file transformer=RegexTransformer,TemplateTransformer field column=rawLine regex=^(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)\t(.*)$ groupNames=field1,field2,field3,field4,field5,field6,field7,field8,field9,field10,field11,field12,field13,field14,field15,field16,field17,field18,field19,field10,field11,field12 / /entity /document /dataConfig Thanks in advance, Mike