[jira] [Comment Edited] (IGNITE-4046) C++: Support DML API
[
https://issues.apache.org/jira/browse/IGNITE-4046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15828043#comment-15828043
]
Igor Sapego edited comment on IGNITE-4046 at 1/18/17 1:31 PM:
--
After some studying, it seems that most flexible and optimized way to introduce
identity resolvers to C++ is to add {{IdentityResolver}} interface, and
introduce {{BynaryType::GetIdentityResolver}} method for user types.
{{IdentityResolver}} should be something like this:
{code}
class IdenntityResolver
{
public:
virtual int32_t GetHashCode(const BinaryObject& obj) = 0;
}
{code}
As you can see, {{BinaryObject}} type used here. It is used for the efficiency
reasons, which Pavel has pointed out. {{BinaryObject}} however is not
implemented currently for C++ client, so if there are no objections I'm going
to add it as a blocking issue for this one.
was (Author: isapego):
After some studying, it seems that most flexible and optimized way to introduce
identity resolvers to C++ is to add {{IdentityResolver}} interface, and
introduce {{BynaryType::GetIdentityResolver}} method for user types.
{{IdentityResolver}} should be something like this:
{code}
class IdenntityResolver
{
virtual int32_t GetHashCode(const BinaryObject& obj) = 0;
}
{code}
As you can see, {{BinaryObject}} type used here. It is used for the efficiency
reasons, which Pavel has pointed out. {{BinaryObject}} however is not
implemented currently for C++ client, so if there are no objections I'm going
to add it as a blocking issue for this one.
> C++: Support DML API
>
>
> Key: IGNITE-4046
> URL: https://issues.apache.org/jira/browse/IGNITE-4046
> Project: Ignite
> Issue Type: Task
> Components: platforms
>Affects Versions: 1.8
>Reporter: Denis Magda
>Assignee: Igor Sapego
> Labels: roadmap
> Fix For: 1.9
>
>
> Ignite's Java component will provide support for DML soon (IGNITE-2294). At
> she same time DML will be supported at the level of ODBC and JDBC drivers.
> As the next step we should include the similar functionality into Ignite.C++
> by doing the following:
> - Implement DML API;
> - Enhance {{query_example.cpp}} by doing INSERTs instead of cache.puts and
> adding UPDATE and DELETE operation examples.
> - Add documentation to Ignite.C++ readme.io covering the feature. Most like
> most of the content can be take from the general documentation when this
> ticket IGNITE-4018 is ready.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Comment Edited] (IGNITE-4046) C++: Support DML API
[
https://issues.apache.org/jira/browse/IGNITE-4046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15825663#comment-15825663
]
Pavel Tupitsyn edited comment on IGNITE-4046 at 1/17/17 8:59 AM:
-
Igor, my main concern is about {{BinaryArrayIdentityResolver}} efficiency: we
do the same thing twice when writing an object to a stream in
{{BinaryWriterImpl::WriteTopObject}} (allocate temp stream, write to temp
stream to get the hash code, then write to the main stream).
Binary writer is on a very hot path, so I think we should get rid of temporary
stream and calculate the hash code using the main stream. Ignite.NET works this
way.
Also please add separate unit tests for {{BinaryArrayIdentityResolver}} and
{{GetDataHashCode}}.
was (Author: ptupitsyn):
Igor, my main concern is about BinaryArrayIdentityResolver efficiency: we do
the same thing twice when writing an object to a stream in
{{BinaryWriterImpl::WriteTopObject}} (allocate temp stream, write to temp
stream to get the hash code, then write to the main stream).
Binary writer is on a very hot path, so I think we should get rid of temporary
stream and calculate the hash code using the main stream. Ignite.NET works this
way.
Also please add separate unit tests for {{BinaryArrayIdentityResolver}} and
{{GetDataHashCode}}.
> C++: Support DML API
>
>
> Key: IGNITE-4046
> URL: https://issues.apache.org/jira/browse/IGNITE-4046
> Project: Ignite
> Issue Type: Task
> Components: platforms
>Affects Versions: 1.8
>Reporter: Denis Magda
>Assignee: Pavel Tupitsyn
> Labels: roadmap
> Fix For: 1.9
>
>
> Ignite's Java component will provide support for DML soon (IGNITE-2294). At
> she same time DML will be supported at the level of ODBC and JDBC drivers.
> As the next step we should include the similar functionality into Ignite.C++
> by doing the following:
> - Implement DML API;
> - Enhance {{query_example.cpp}} by doing INSERTs instead of cache.puts and
> adding UPDATE and DELETE operation examples.
> - Add documentation to Ignite.C++ readme.io covering the feature. Most like
> most of the content can be take from the general documentation when this
> ticket IGNITE-4018 is ready.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Comment Edited] (IGNITE-4046) C++: Support DML API
[
https://issues.apache.org/jira/browse/IGNITE-4046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15825663#comment-15825663
]
Pavel Tupitsyn edited comment on IGNITE-4046 at 1/17/17 8:59 AM:
-
Igor, my main concern is about BinaryArrayIdentityResolver efficiency: we do
the same thing twice when writing an object to a stream in
{{BinaryWriterImpl::WriteTopObject}} (allocate temp stream, write to temp
stream to get the hash code, then write to the main stream).
Binary writer is on a very hot path, so I think we should get rid of temporary
stream and calculate the hash code using the main stream. Ignite.NET works this
way.
Also please add separate unit tests for {{BinaryArrayIdentityResolver}} and
{{GetDataHashCode}}.
was (Author: ptupitsyn):
Igor, my main concern is about BinaryArrayIdentityResolver efficiency: we do
the same thing twice when writing an object to a stream in
{{BinaryWriterImpl::WriteTopObject}} (allocate temp stream, write to temp
stream to get the hash code, then write to the main stream).
Binary writer is on a very hot path, so I think we should get rid of temporary
stream and calculate the hash code using the main stream. Ignite.NET works this
way.
> C++: Support DML API
>
>
> Key: IGNITE-4046
> URL: https://issues.apache.org/jira/browse/IGNITE-4046
> Project: Ignite
> Issue Type: Task
> Components: platforms
>Affects Versions: 1.8
>Reporter: Denis Magda
>Assignee: Pavel Tupitsyn
> Labels: roadmap
> Fix For: 1.9
>
>
> Ignite's Java component will provide support for DML soon (IGNITE-2294). At
> she same time DML will be supported at the level of ODBC and JDBC drivers.
> As the next step we should include the similar functionality into Ignite.C++
> by doing the following:
> - Implement DML API;
> - Enhance {{query_example.cpp}} by doing INSERTs instead of cache.puts and
> adding UPDATE and DELETE operation examples.
> - Add documentation to Ignite.C++ readme.io covering the feature. Most like
> most of the content can be take from the general documentation when this
> ticket IGNITE-4018 is ready.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Comment Edited] (IGNITE-4046) C++: Support DML API
[
https://issues.apache.org/jira/browse/IGNITE-4046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15728740#comment-15728740
]
Igor Sapego edited comment on IGNITE-4046 at 12/7/16 1:22 PM:
--
Implemented algorithm that matches {{BinaryArrayIdentityResolver}}'s. When I've
tried to implement {{BinaryFieldIdentityResolver}} I've found out that in Java
it calculates hash by applying hash function to set of the fields hashes. Field
hashes in their turn calculated using Java-specific algorithms and Java data
types internal structure. I don't see a way to implement the algorithm, which
is going to be able to the same hash in C++ or any other platform.
It is obvious that the portable hashing algorithm should only use binary format
to generate object hash. {{BinaryFieldIdentityResolver}} is not portable as for
now.
was (Author: isapego):
Implemented algorithm that matches {{BinaryArrayIdentityResolver}}'s. When I've
tried to implement {{BinaryFieldIdentityResolver}} I've found out that in Java
it calculates hash by applying hash function to set of the fields hashes. Field
hashes in their turn calculated using Java-specific algorithms and Java data
types internal structure. I don't see a way to implement the algorithm, which
is going to be able to the same hash in C++ or any other platform.
> C++: Support DML API
>
>
> Key: IGNITE-4046
> URL: https://issues.apache.org/jira/browse/IGNITE-4046
> Project: Ignite
> Issue Type: Task
> Components: platforms
>Reporter: Denis Magda
>Assignee: Igor Sapego
> Labels: roadmap
> Fix For: 2.0
>
>
> Ignite's Java component will provide support for DML soon (IGNITE-2294). At
> she same time DML will be supported at the level of ODBC and JDBC drivers.
> As the next step we should include the similar functionality into Ignite.C++
> by doing the following:
> - Implement DML API;
> - Enhance {{query_example.cpp}} by doing INSERTs instead of cache.puts and
> adding UPDATE and DELETE operation examples.
> - Add documentation to Ignite.C++ readme.io covering the feature. Most like
> most of the content can be take from the general documentation when this
> ticket IGNITE-4018 is ready.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
