Date   

Re: JanusGraph database cache on distributed setup

Boxuan Li
 

Hi Wasantha,

In your example, it looks like you didn't commit your transaction on JG instance A. Uncommitted changes are only visible to the local transaction on the local instance. Can you try committing it first on A and then query on B?

Best,
Boxuan


Re: JanusGraph database cache on distributed setup

washerath@...
 

Hi Boxuan,

I was able to change ExpirationKCVSCache class to persist the cache on Redis DB,

But i could still see some data anomaly between two JG instances. As example when i change a property of a vertex from one JG server [ g.V(40964200).property('data', 'some_other_value')

JG instance A
JG instance A

it does not reflect on other JG instance. 

JG instance B
JG instance B

When debuging the flow we could identify that when we triggering a vertex property modification, it gets persists on guava cache using  GuavaVertexCache add method and when retrieving it reads data using  get method on same class. This could be the reason for above observation.

Feels like we might need to do modifications on transaction-wise cache as well. Correct me if i am missing something here and happy to contribute the implementation to the community once this done.

Thanks

Wasantha


Re: JanusGraph database cache on distributed setup

Boxuan Li
 

Hi Wasantha,


It's great to see that you have made some progress. If possible, it would be awesome if you could contribute your implementation to the community!

Yes, modifying `ExpirationKCVSCache`  is enough. `GuavaSubqueryCache` and `GuavaVertexCache` and transaction-wise cache, so you don't want to make them global. `RelationQueryCache` and `SchemaCache` are graph-wise cache, you could make them global, but not necessary since they only store schema rather than real data - actually I would recommend not doing so because JanusGraph already has a mechanism of evicting stale schema cache.

Best,
Boxuan


Re: JanusGraph database cache on distributed setup

washerath@...
 

Hi Boxuan,

I am evaluating the approach of rewriting ExpirationKCVSCache as suggested. There i could replace existing guava cache implementation to connect with remote Redis db.  So that Redis DB will be act as centralized cache which can connects with all other JG instances.

While going through the JG source it could find same guava cache implementation (cachebuilder = CacheBuilder.newBuilder()) uses on several other places. Eg . GuavaSubqueryCache, GuavaVertexCache, ...

Will it be sufficient to have modification only on ExpirationKCVSCache or do we need to look for modification on several other places as well ?

Thanks

Wasantha


Re: Removed graphs still open in muti node cluster

hadoopmarc@...
 

Hi Lixu,

JanusGraph-0.6.0 had various changes to the ConfiguredGraphFactory which might have solved your issue:

https://github.com/JanusGraph/janusgraph/issues/2236
https://github.com/JanusGraph/janusgraph/blob/v0.5.3/janusgraph-core/src/main/java/org/janusgraph/core/ConfiguredGraphFactory.java
https://github.com/JanusGraph/janusgraph/blob/v0.6.1/janusgraph-core/src/main/java/org/janusgraph/core/ConfiguredGraphFactory.java

Can you recheck with version 0.6.1?

BTW, the release notes of v0.6.0 form an impressive list! Merely reading it takes minutes.

Best wishes,

Marc


Re: Preserve IDs when importing graphml

hadoopmarc@...
 

Hi Laura,

No answer but some relevant search results:

https://groups.google.com/g/gremlin-users/c/jUBuhhKuf0M/m/kiKMY0eHAwAJ
The  graph.set-vertex-id property at: https://docs.janusgraph.org/configs/configuration-reference/#graph

In general, when working with JanusGraph, it is better to first transform the input graphml and make the id into a property.

Best wishes,   Marc


Preserve IDs when importing graphml

Laura Morales <lauretas@...>
 

I think I've read once that it's possible to preserve the IDs when importing graphml data. Unfortunately, I cannot remember where I read that. All my IDs are integers.
How do I do that?


Re: Importing a schema

hadoopmarc@...
 

Hi Laura,

JanusGraph only allows to configure a custom SchemaMaker with the schema.default property. Googling on SchemaMaker hits some (unmaintained?) projects that could help:

https://github.com/graph-lab/janusgraph-schema-manager

Best wishes,   Marc


Importing a schema

Laura Morales <lauretas@...>
 

Is there a way to import a schema, instead of creating it with a script?
For example importing this file:

<?xml version='1.0' ?>
<graphml xmlns='http://graphml.graphdrawing.org/xmlns'>
<key id='type' for='node' attr.name='type' attr.type='string'></key>
<key id='code' for='node' attr.name='code' attr.type='string'></key>
<key id='icao' for='node' attr.name='icao' attr.type='string'></key>
<key id='desc' for='node' attr.name='desc' attr.type='string'></key>
<key id='region' for='node' attr.name='region' attr.type='string'></key>
<key id='runways' for='node' attr.name='runways' attr.type='int'></key>
<key id='longest' for='node' attr.name='longest' attr.type='int'></key>
<key id='elev' for='node' attr.name='elev' attr.type='int'></key>
<key id='country' for='node' attr.name='country' attr.type='string'></key>
<key id='city' for='node' attr.name='city' attr.type='string'></key>
<key id='lat' for='node' attr.name='lat' attr.type='double'></key>
<key id='lon' for='node' attr.name='lon' attr.type='double'></key>
<key id='dist' for='edge' attr.name='dist' attr.type='int'></key>
<key id='labelV' for='node' attr.name='labelV' attr.type='string'></key>
<key id='labelE' for='edge' attr.name='labelE' attr.type='string'></key>
</graphml>

like this: graph.io(graphml()).readGraph("file.graphml")

is there a way to make Janus create the schema from the XML file? This would be very convenient because it means I don't have to write groovy scripts for creating the schema.


Re: JanusGraph database cache on distributed setup

Boxuan Li
 

Hi Wasantha,

A centralized cache is a good idea in many use cases. What you could do is to maintain a centralized cache by yourself. This, however, requires some changes to your application code (e.g. your app might need to do a look up in cache and then query JanusGraph). A more advanced approach is to rewrite ExpirationKCVSCache (https://javadoc.io/doc/org.janusgraph/janusgraph-core/latest/org/janusgraph/diskstorage/keycolumnvalue/cache/ExpirationKCVSCache.html) by yourself and let it store cache in a centralized cache rather than the local cache. Then, the db.cache feature should still work except that the cache is synced across JanusGraph instances.

Best,
Boxuan

On Feb 10, 2022, at 10:59 PM, washerath@... wrote:

Actually the concern is with db.cache feature.

Once we enable the db.cache, what ever the modification done for particular vertex only visible for that JG instance untill the cache expires. So if we have multiple JG instances, the modifications done from one instance does not reflect on other immediately. 

If we can have centralized cache which syncs up with all JG instances this can be avoided.

Thanks, Wasantha


Re: JanusGraph database cache on distributed setup

washerath@...
 

Actually the concern is with db.cache feature.

Once we enable the db.cache, what ever the modification done for particular vertex only visible for that JG instance untill the cache expires. So if we have multiple JG instances, the modifications done from one instance does not reflect on other immediately. 

If we can have centralized cache which syncs up with all JG instances this can be avoided.

Thanks, Wasantha


can we dynamically create multiple graphs with customized schema files

Yingjie Li
 

Hello,

Currently we run gremlin.sh with  customized schema file in groovy  (contains backend config, graph name, cache size as well as property key, vertex/edge index)  to initialize graph. It seems that we have to restart the server to make sure that the graph is accessible. 

Is there a way to dynamically create multiple graphs with customized schema files  without the need to restart the server?  


Removed graphs still open in muti node cluster

lixu
 

Hi, I'm using JnausGraphManager and JanusGraphWsAndHttpChannelizer to manage dynamic graph operating.
When dropping graph in multi-nodes cluster, removed graphs only closed in specific node, the removed graphs are still open in other nodes.
 
Version: 0.5.3 
Storage Backend: Hbase
Mixed Index Backend: elasticsearch
Expected Behavior: all nodes in cluster should close the removed graphs
Current Behavior: only the node executing dropping script close the removed graphs
 
Related codes:
JanusGraphManager:
 
    private class GremlinExecutorGraphBinder implements Runnable {
        final JanusGraphManager graphManager;
        final GremlinExecutor gremlinExecutor;
 
        public GremlinExecutorGraphBinder(JanusGraphManager graphManager, GremlinExecutor gremlinExecutor) {
            this.graphManager = graphManager;
            this.gremlinExecutor = gremlinExecutor;
        }
 
        @Override
        public void run() {
            ConfiguredGraphFactory.getGraphNames().forEach(it -> {
                try {
                    final Graph graph = ConfiguredGraphFactory.open(it);
                    updateTraversalSource(it, graph, this.gremlinExecutor, this.graphManager);
                } catch (Exception e) {
                    // cannot open graph, do nothing
                    log.error(String.format("Failed to open graph %s with the following error:\n %s.\n" +
                    "Thus, it and its traversal will not be bound on this server.", it, e.toString()));
                }
            });
        }
    }
 
In above codes, run() method will get all the graph names from hbase, so it could find all the added graphs, 
but those graphs removed from the other nodes are still open in current node.
 
 
 
 


Re: Forcing Janusgraph to use indices when performing traversal with Union step

Boxuan Li
 

The approach you proposed should work as good/bad as your original single query in theory. I would be surprised if this approach works better than your original single query, but if so, please let me know.

The optimal approach depends on your data. Since each of your indexes only covers a single index, there are two strategies in general:

1. Leverage one index to return results for one condition, and do in-memory filtering for the other condition. Your original query does this. This is useful in some scenarios (e.g. one index returns many more results than another index).

2. Leverage both indexes and do in-memory intersection to return results that satisfy both conditions. This might be something you want, and it is useful in some scenarios (e.g. both indices return roughly the same results and there is not much overlapping between two sets of results). If this is what you want, try rewriting your query as g.V().has("indexed-prop1", "value1").or(__.has("indexed-prop2", "value2"), __.has("indexed-prop3", "value3")). Essentially, change the union step into "or" step.

Hope this helps!

Best,
Boxuan


Re: Forcing Janusgraph to use indices when performing traversal with Union step

brad@...
 

Thank you for your reply.

We can look into migrating to version 0.6 (or later)...

There exists an index for indexed-prop1, indexed-prop2, and indexed-prop3, but there are no multi-column indices (e.g., indexed-prop1 and indexed-prop2).  In our application, the columns that might be used for the search are specified by the user, and we wouldn't have any way of knowing in advance what combination of columns they might ask for.  The only thing that we can do now is to ensure that all of the columns that are used in a 'has' step are indexed.  

One approach that I thought might be an improvement is to perform the first traversal (i.e., " g.V().has('indexed-prop1', 'value1')") separately, collect the set of vertices which satisfy this query, then perform an additional query for each 'has' step that is currently in the 'union' step.  So, for example, we run g.V().has('indexed-prop1', 'value1') initially, and it returns 3 vertices (V1, V2, and V3).  We then run a traversal for indexed-prop2:   "g.V(V1, V2, V3).has('indexed-prop2', 'value2')", which returns a subset of the vertices returned by the first traversal.  Then, run another traversal, this time:   "g.V(V1, V2, V3).has('indexed-prop3', 'value3')", again returning a subset of the vertices.  Finally, figure out (programatically, without executing another traversal) the union of the vertices returned by the last 2 traversals.  This is an inelegant and brute-force technique which would probably work, but I would rather do this in a single traversal, and I haven't been able to figure out how to do this.  Can you recommend an approach for doing this kind of action in a single traversal?

Thanks,
Brad


Re: Forcing Janusgraph to use indices when performing traversal with Union step

Boxuan Li
 

Hi Brad,

I can see that your traversal is using the index index_ten1_apm_0 from the following snippet:

  backend-query                                                        3                      21.484
    \\_query=index_ten1_apm_0:[(ten1.apm.idx.type_id = i_javaServiceInstance)](4000):index_ten1_apm_0
    \\_limit=4000

Then, JanusGraph uses in-memory filtering to check whether the results returned by the index_ten1_apm_0 satisfy your predicates has('indexed-prop2', 'value2’) or has('indexed-prop3', ‘value3’).

Do you have an index that contains both the field “indexed-prop1” and “indexed-prop2”, and an index that contains both the field “indexed-prop1” and “indexed-prop3”? If so, try replacing “union” step with “or” step. If not, then you could try creating those indexes, otherwise there is no optimization that JanusGraph can do.

Btw it looks like you are using a JanusGraph version < 0.6, which is out of maintenance. The same query should run moderately faster in the latest version due to a couple of optimizations.

Best,
Boxuan

On Feb 6, 2022, at 3:14 PM, brad@... wrote:

TRAVERSAL:
[GraphStep(vertex,[]), HasStep([~label.eq(ten1.apm.version), ten1.apm.idx.type_id.eq(i_javaServiceInstance)]), UnionStep([[HasStep([ten1.apm.idx.display_name.eq(JavaServiceInstance3)]), EndStep], [HasStep([ten1.apm.idx.str4.eq(998)]), EndStep]]), RangeGlobalStep(0,2147483647), HasStep([ten1.apm.idx.discovery_ts.lte(1643313164000), ten1.apm.idx.last_seen_ts.gt(0)]), OrderGlobalStep([[value(ten1.apm.idx.last_seen_ts), desc]]), RangeGlobalStep(0,1000), GroupStep(value(uid),[FoldStep, OrderLocalStep([[value(ten1.apm.idx.last_seen_ts), desc]]), UnfoldStep, RangeGlobalStep(0,1)]), LambdaFlatMapStep(lambda), RangeGlobalStep(0,2500), PropertyMapStep([uid, ten1.apm.idx.display_name, ten1.apm.idx.last_seen_ts, ten1.apm.idx.discovery_ts, typ],value)]
Traversal Metrics
Step                                                               Count  Traversers       Time (ms)    % Dur
=============================================================================================================
JanusGraphStep([],[~label.eq(ten1.apm.version),...                     3           3          14.381    54.44
    \\_condition=(~label = ten1.apm.version AND ten1.apm.idx.type_id = i_javaServiceInstance)
    \\_orders=[]
    \\_isFitted=true
    \\_isOrdered=true
    \\_query=[(ten1.apm.idx.type_id = i_javaServiceInstance)](4000):index_ten1_apm_0
    \\_index=index_ten1_apm_0
    \\_index_impl=search
  optimization                                                                                 0.011
  optimization                                                                                 0.381
  backend-query                                                        3                      21.484
    \\_query=index_ten1_apm_0:[(ten1.apm.idx.type_id = i_javaServiceInstance)](4000):index_ten1_apm_0
    \\_limit=4000
UnionStep([[HasStep([ten1.apm.idx.display_name....                     2           2          10.005    37.87
  HasStep([ten1.apm.idx.display_name.eq(JavaSer...                     1           1           0.113
  EndStep                                                              1           1           0.076
  HasStep([ten1.apm.idx.str4.eq(998)])                                 1           1           0.208
  EndStep                                                              1           1           0.169
RangeGlobalStep(0,2147483647)                                          2           2           0.177     0.67
HasStep([ten1.apm.idx.discovery_ts.lte(16433131...                     2           2           0.616     2.33
OrderGlobalStep([[value(ten1.apm.idx.last_seen_...                     2           2           0.281     1.06
RangeGlobalStep(0,1000)                                                2           2           0.115     0.44
GroupStep(value(uid),[FoldStep, OrderLocalStep(...                     1           1           0.323     1.22
LambdaFlatMapStep(lambda)                                              2           2           0.082     0.31
RangeGlobalStep(0,2500)                                                2           2           0.062     0.23
PropertyMapStep([uid, ten1.apm.idx.display_name...                     2           2           0.373     1.41
                                            >TOTAL                     -           -          26.417        -
 


Re: Forcing Janusgraph to use indices when performing traversal with Union step

brad@...
 

TRAVERSAL:
[GraphStep(vertex,[]), HasStep([~label.eq(ten1.apm.version), ten1.apm.idx.type_id.eq(i_javaServiceInstance)]), UnionStep([[HasStep([ten1.apm.idx.display_name.eq(JavaServiceInstance3)]), EndStep], [HasStep([ten1.apm.idx.str4.eq(998)]), EndStep]]), RangeGlobalStep(0,2147483647), HasStep([ten1.apm.idx.discovery_ts.lte(1643313164000), ten1.apm.idx.last_seen_ts.gt(0)]), OrderGlobalStep([[value(ten1.apm.idx.last_seen_ts), desc]]), RangeGlobalStep(0,1000), GroupStep(value(uid),[FoldStep, OrderLocalStep([[value(ten1.apm.idx.last_seen_ts), desc]]), UnfoldStep, RangeGlobalStep(0,1)]), LambdaFlatMapStep(lambda), RangeGlobalStep(0,2500), PropertyMapStep([uid, ten1.apm.idx.display_name, ten1.apm.idx.last_seen_ts, ten1.apm.idx.discovery_ts, typ],value)]
Traversal Metrics
Step                                                               Count  Traversers       Time (ms)    % Dur
=============================================================================================================
JanusGraphStep([],[~label.eq(ten1.apm.version),...                     3           3          14.381    54.44
    \\_condition=(~label = ten1.apm.version AND ten1.apm.idx.type_id = i_javaServiceInstance)
    \\_orders=[]
    \\_isFitted=true
    \\_isOrdered=true
    \\_query=[(ten1.apm.idx.type_id = i_javaServiceInstance)](4000):index_ten1_apm_0
    \\_index=index_ten1_apm_0
    \\_index_impl=search
  optimization                                                                                 0.011
  optimization                                                                                 0.381
  backend-query                                                        3                      21.484
    \\_query=index_ten1_apm_0:[(ten1.apm.idx.type_id = i_javaServiceInstance)](4000):index_ten1_apm_0
    \\_limit=4000
UnionStep([[HasStep([ten1.apm.idx.display_name....                     2           2          10.005    37.87
  HasStep([ten1.apm.idx.display_name.eq(JavaSer...                     1           1           0.113
  EndStep                                                              1           1           0.076
  HasStep([ten1.apm.idx.str4.eq(998)])                                 1           1           0.208
  EndStep                                                              1           1           0.169
RangeGlobalStep(0,2147483647)                                          2           2           0.177     0.67
HasStep([ten1.apm.idx.discovery_ts.lte(16433131...                     2           2           0.616     2.33
OrderGlobalStep([[value(ten1.apm.idx.last_seen_...                     2           2           0.281     1.06
RangeGlobalStep(0,1000)                                                2           2           0.115     0.44
GroupStep(value(uid),[FoldStep, OrderLocalStep(...                     1           1           0.323     1.22
LambdaFlatMapStep(lambda)                                              2           2           0.082     0.31
RangeGlobalStep(0,2500)                                                2           2           0.062     0.23
PropertyMapStep([uid, ten1.apm.idx.display_name...                     2           2           0.373     1.41
                                            >TOTAL                     -           -          26.417        -
 


Re: dynamic graphics, limits and global index

Matthew Nguyen <nguyenm9@...>
 

Thanks Marc.  Currently triplestore/LPG is on hold awaiting streaming incidental edge queries in order to play some more.  Hoping we will see a day when LPG/3store harmonize.  


-----Original Message-----
From: hadoopmarc@...
To: janusgraph-users@...
Sent: Sun, Feb 6, 2022 1:56 pm
Subject: Re: [janusgraph-users] dynamic graphics, limits and global index

Hi Matt,

Adding to what I stated above about independent composite indices for separate graphs on the same storage backend, the issue turns out to more nuanced for mixed indices on an indexing bakend, see the recent question:

https://lists.lfaidata.foundation/g/janusgraph-users/topic/88879391

I though it useful to add it to this thread too.

Marc

PS Good to hear that JanusGraph can possibly support your usecase!


Re: Forcing Janusgraph to use indices when performing traversal with Union step

Boxuan Li
 

Hi Brad,

Can you post the profile result of both queries? You can retrieve profile results by adding `.profile()` to the end of your query.

Best,
Boxuan


Re: JanusGraph database cache on distributed setup

hadoopmarc@...
 

How fast is immediately?  A well dimensioned cassandra or scylladb cluster (with its own block cache!) should be able to serve requests at the ms level.

https://www.scylladb.com/2017/09/18/scylla-2-0-workload-conditioning/

You only run into trouble with queries that ask for tens or hundreds of vertices, but you can ask if it is reasonable to be realtime at the ms level for such large queries.

Best wishes,     Marc

241 - 260 of 6656