Hello,

Does anybody know how ids are generated in a distributed environment? 

I am trying to figure out how JanusGraph generates identifiers for new vertices. As I understand each vertex has a 64 bit identifier which determines in which partition the vertex is stored.

The only problem is that there is no any information about how those IDs are generated. Is there a possibility for duplicates when several instances are creating new vertices simultaneously? Is there any possibility to change the identifier of the vertex when it is persisted already? Are instances talking to each other to synchronize identifiers?

Best regards,

Alexandr

Multiple instances can not duplicate vertex id, as there is a concept of  unique id block assignment to each instance in case of unused this is not reassigned to other instance. Check property configuration: 

ids.block-size 

~

On Tuesday, December 26, 2017 at 2:58:29 AM UTC+5:30, Alexandr Porunov wrote:

Hello,

Does anybody know how ids are generated in a distributed environment? 

I am trying to figure out how JanusGraph generates identifiers for new vertices. As I understand each vertex has a 64 bit identifier which determines in which partition the vertex is stored.

The only problem is that there is no any information about how those IDs are generated. Is there a possibility for duplicates when several instances are creating new vertices simultaneously? Is there any possibility to change the identifier of the vertex when it is persisted already? Are instances talking to each other to synchronize identifiers?

Best regards,

Alexandr

Ankur,

Thank you very much for pointing out. I tried to investigate more deeply into the id allocation process, but I am still confused a little bit.

I would be very grateful If you can answer one more question.

I still don't understand how different JanusGraph instances allocate id blocks. What if two different JanusGraph instances allocate the same id block? For example, if we are using two multi datacenter Cassandra clusters (which has an asynchronous syncing) we cannot guarantee consistency (because Cassandra is eventually consistent database). So, there is a probability that two different JanusGraph instances allocate the same id block (each of the instances would think that only it allocated the block). Is there any protection in JanusGraph against such situation? What would happen in such scenario?

Best regards,

Alexandr

On Tuesday, December 26, 2017 at 6:32:12 AM UTC+2, Ankur Goel wrote:

Multiple instances can not duplicate vertex id, as there is a concept of  unique id block assignment to each instance in case of unused this is not reassigned to other instance. Check property configuration: 

ids.block-size 

~

On Tuesday, December 26, 2017 at 2:58:29 AM UTC+5:30, Alexandr Porunov wrote:

Hello,

Does anybody know how ids are generated in a distributed environment? 

I am trying to figure out how JanusGraph generates identifiers for new vertices. As I understand each vertex has a 64 bit identifier which determines in which partition the vertex is stored.

The only problem is that there is no any information about how those IDs are generated. Is there a possibility for duplicates when several instances are creating new vertices simultaneously? Is there any possibility to change the identifier of the vertex when it is persisted already? Are instances talking to each other to synchronize identifiers?

Best regards,

Alexandr

Hello.

It utilizes transactions (JanusGraph-level) internally:

* https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/diskstorage/idmanagement/ConsistentKeyIDAuthority.java#L260
* https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/diskstorage/idmanagement/ConsistentKeyIDAuthority.java#L285
*https://github.com/JanusGraph/janusgraph/blob/1c864333e709a4445c049b051855d726decb56d8/janusgraph-core/src/main/java/org/janusgraph/diskstorage/util/BackendOperation.java#L142

See configuration options for tuning: https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/graphdb/configuration/GraphDatabaseConfiguration.java#L700

Also retries in case id block is already allocated in the backend. Using multi-dc setup with default options is probably safe (?) - as far as I can tell from https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-cql/src/main/java/org/janusgraph/diskstorage/cql/CQLConfigOptions.java - by default system operations (transactions too) will utilize QUORUM consistency in a cross-DC manner. You will have to use NetworkTopologyStrategy in JanusGraph configuration (SimpleStrategy by default for Janus keyspaces).

Best regards,
Evgenii Ignatev.

On 6/21/2020 6:36 AM, Chen Wu wrote:

Hi, any updates on this?  I'm thinking about this question too

在 2017年12月30日星期六 UTC+8上午3:56:51，Alexandr Porunov写道：

Ankur,

Thank you very much for pointing out. I tried to investigate more deeply into the id allocation process, but I am still confused a little bit.

I would be very grateful If you can answer one more question.

I still don't understand how different JanusGraph instances allocate id blocks. What if two different JanusGraph instances allocate the same id block? For example, if we are using two multi datacenter Cassandra clusters (which has an asynchronous syncing) we cannot guarantee consistency (because Cassandra is eventually consistent database). So, there is a probability that two different JanusGraph instances allocate the same id block (each of the instances would think that only it allocated the block). Is there any protection in JanusGraph against such situation? What would happen in such scenario?

Best regards,

Alexandr

On Tuesday, December 26, 2017 at 6:32:12 AM UTC+2, Ankur Goel wrote:

Multiple instances can not duplicate vertex id, as there is a concept of  unique id block assignment to each instance in case of unused this is not reassigned to other instance. Check property configuration: 

ids.block-size 

~

On Tuesday, December 26, 2017 at 2:58:29 AM UTC+5:30, Alexandr Porunov wrote:

Hello,

Does anybody know how ids are generated in a distributed environment? 

I am trying to figure out how JanusGraph generates identifiers for new vertices. As I understand each vertex has a 64 bit identifier which determines in which partition the vertex is stored.

The only problem is that there is no any information about how those IDs are generated. Is there a possibility for duplicates when several instances are creating new vertices simultaneously? Is there any possibility to change the identifier of the vertex when it is persisted already? Are instances talking to each other to synchronize identifiers?

Best regards,

Alexandr

--
You received this message because you are subscribed to the Google Groups "JanusGraph users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to janusgra...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/janusgraph-users/a1b77ea0-bf53-4303-ba6a-44fe9d1def30o%40googlegroups.com.

-- 
Best regards,
Evgeniy Ignatiev.

Thank you for your reply.

I've also read the code in ConsistentKeyIdAuthority.java, what I'm confusing now is as follows:

in https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/diskstorage/idmanagement/ConsistentKeyIDAuthority.java#L307,

when one claim(claim1) is beat by another claim(claim2) on block [start, end], does claim1 retry calim on  block [start, end], or it trys to claim on another block [start_new, end_new]? I think it should be the latter case but I can't find the code to claim another block.

It's very helpful if you could give some hints, thanks.

在 2020年6月21日星期日 UTC+8下午8:10:27，Evgeniy Ignatiev写道：

Hello.

It utilizes transactions (JanusGraph-level) internally:

* https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/diskstorage/idmanagement/ConsistentKeyIDAuthority.java#L260
* https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/diskstorage/idmanagement/ConsistentKeyIDAuthority.java#L285
*https://github.com/JanusGraph/janusgraph/blob/1c864333e709a4445c049b051855d726decb56d8/janusgraph-core/src/main/java/org/janusgraph/diskstorage/util/BackendOperation.java#L142

See configuration options for tuning: https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-core/src/main/java/org/janusgraph/graphdb/configuration/GraphDatabaseConfiguration.java#L700

Also retries in case id block is already allocated in the backend. Using multi-dc setup with default options is probably safe (?) - as far as I can tell from https://github.com/JanusGraph/janusgraph/blob/master/janusgraph-cql/src/main/java/org/janusgraph/diskstorage/cql/CQLConfigOptions.java - by default system operations (transactions too) will utilize QUORUM consistency in a cross-DC manner. You will have to use NetworkTopologyStrategy in JanusGraph configuration (SimpleStrategy by default for Janus keyspaces).

Best regards,
Evgenii Ignatev.

On 6/21/2020 6:36 AM, Chen Wu wrote:

Hi, any updates on this?  I'm thinking about this question too

在 2017年12月30日星期六 UTC+8上午3:56:51，Alexandr Porunov写道：

Ankur,

Thank you very much for pointing out. I tried to investigate more deeply into the id allocation process, but I am still confused a little bit.

I would be very grateful If you can answer one more question.

I still don't understand how different JanusGraph instances allocate id blocks. What if two different JanusGraph instances allocate the same id block? For example, if we are using two multi datacenter Cassandra clusters (which has an asynchronous syncing) we cannot guarantee consistency (because Cassandra is eventually consistent database). So, there is a probability that two different JanusGraph instances allocate the same id block (each of the instances would think that only it allocated the block). Is there any protection in JanusGraph against such situation? What would happen in such scenario?

Best regards,

Alexandr

On Tuesday, December 26, 2017 at 6:32:12 AM UTC+2, Ankur Goel wrote:

Multiple instances can not duplicate vertex id, as there is a concept of  unique id block assignment to each instance in case of unused this is not reassigned to other instance. Check property configuration: 

ids.block-size 

~

On Tuesday, December 26, 2017 at 2:58:29 AM UTC+5:30, Alexandr Porunov wrote:

Hello,

Does anybody know how ids are generated in a distributed environment? 

I am trying to figure out how JanusGraph generates identifiers for new vertices. As I understand each vertex has a 64 bit identifier which determines in which partition the vertex is stored.

The only problem is that there is no any information about how those IDs are generated. Is there a possibility for duplicates when several instances are creating new vertices simultaneously? Is there any possibility to change the identifier of the vertex when it is persisted already? Are instances talking to each other to synchronize identifiers?

Best regards,

Alexandr

--
You received this message because you are subscribed to the Google Groups "JanusGraph users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to janusgra...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/janusgraph-users/a1b77ea0-bf53-4303-ba6a-44fe9d1def30o%40googlegroups.com.

-- 
Best regards,
Evgeniy Ignatiev.

--
You received this message because you are subscribed to the Google Groups "JanusGraph users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to janusgra...@....
To view this discussion on the web visit https://groups.google.com/d/msgid/janusgraph-users/0286e8ee-54e1-4a22-94d7-324dea7a7aa0o%40googlegroups.com.