Upgrading from 0.6 to 1.0.0-20230626 caused indexes to disappear

Hi all, I tried to upgrade from 0.6 to 1.0.0-20230626 yesterday, and I ran into some issues. All of the existing indexes was gone after the upgrade. My setup is cql-es, using cassandra 3.11.6 and ES 7.16.2. If I recall correctly, the graph used to contain around 10 CompositeIndexes. Is this a known issue? Should I simply rebuild the indexes? I've posted the (reduced) output of mgmt.printSchema() below.
---------------------------------------------------------------------------------------------------
Graph Index (Vertex) | Type | Unique | Backing | Key: Status |
---------------------------------------------------------------------------------------------------
Graph_Name_Index | Composite | true | internalindex | graph.graphname: ENABLED |
Template_Index | Composite | false | internalindex | Template_Configuration: ENABLED |
Created_Using_Template_Index | Composite | false | internalindex | Created_Using_Template: ENABLED |
---------------------------------------------------------------------------------------------------
Graph Index (Edge) | Type | Unique | Backing | Key: Status |
---------------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------------
Relation Index (VCI) | Type | Direction | Sort Key | Order | Status |
---------------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------------
Graph Index (Vertex) | Type | Unique | Backing | Key: Status |
---------------------------------------------------------------------------------------------------
Graph_Name_Index | Composite | true | internalindex | graph.graphname: ENABLED |
Template_Index | Composite | false | internalindex | Template_Configuration: ENABLED |
Created_Using_Template_Index | Composite | false | internalindex | Created_Using_Template: ENABLED |
---------------------------------------------------------------------------------------------------
Graph Index (Edge) | Type | Unique | Backing | Key: Status |
---------------------------------------------------------------------------------------------------
---------------------------------------------------------------------------------------------------
Relation Index (VCI) | Type | Direction | Sort Key | Order | Status |
---------------------------------------------------------------------------------------------------
5 Replies
slygren
slygren17mo ago
Probably some of the same I'm experiencing here! On 1.0 RC2 I've created indexes and can verify them being ENABLEd and used (checked .profile()) however - doing a simple g.V().has('someindex', 'somevalue').count() turns out 0 as if the 'someindex' property does not seem to exist (while it does).
Florian Hockmann
Florian Hockmann17mo ago
The indexes printed by mgmt.printSchema() are all from the ConfigurationManagementGraph. This looks like you are printing the schema from the management graph, but I guess you have probably created the schema for one graph that ws created with the template configuration. Is that true? In that case you also need to open the management system for that graph. You can do that like this:
graph2 = ConfiguredGraphFactory.open("[graphname]");
mgmt = graph2.openManagement()
graph2 = ConfiguredGraphFactory.open("[graphname]");
mgmt = graph2.openManagement()
@slygren I think your problem is unrelated to this. Or are you also using the ConfigurationManagementGraph and see a similar output when you execute mgmt.printSchema()? If not, then please create a separate thread for your question / problem. Otherwise, we mix up two different topics here which makes it hard to keep track of them
slygren
slygren17mo ago
@florianhockmann - you're correct - my issue seems to be related to transactions and consistency using ScyllaDB as a backend
rydbirkjr
rydbirkjr17mo ago
@florianhockmann , I think you're right regarding the schema printing. However, the queries stopped using the indexes nonetheless, so I'm unsure why they got disabled. Just for future reference, what I ended up doing was to get rid of the ConfigurationManagementGraph and rebuilding the graph from source. ConfigurationManagementGraph wasn't needed anymore in our setup and we switched to only having the default graph as defined in gremlin-server.yaml. Indexes looks fine now.
4j4y.
4j4y.9mo ago
@rydbirkjr What were the upgrade steps you took? Was it just docker image bump up?
Want results from more Discord servers?
Add your server