Neo4j bulk import fails: No label was specified for the node index in 'id:ID'

[c4jquick]

Using csvs and bash script exported from ftm export-neo4j-bulk i am getting errors when importing into Neo4j Desktop

any ideas? @pudo

versions

macOS 26.3.1 (25D2128)
Neo4j Desktop 2 Version 2.1.3 (2.1.3)
followthemoney==4.3.3

java env from dbms in desktop instance:
openjdk version "21.0.8" 2025-07-15 LTS
OpenJDK Runtime Environment Zulu21.44+17-CA (build 21.0.8+9-LTS)
OpenJDK 64-Bit Server VM Zulu21.44+17-CA (build 21.0.8+9-LTS, mixed mode, sharing)
neo4j-admin 2025.11.2

export script

ftm export-neo4j-bulk works fine, creates folder w/ csvs/script

the generated neo4j_import.sh script has the following problem w/ neo4j im using. think this is for an older version but easily fixable. i can make a pr if ur interested in using this cmd format

# works
neo4j-admin database import incremental <db name> --force ...
# original 
neo4j-admin import --database=graph.db ...

fully fixed script

desktop="/Users/dardanus/Library/Application Support/neo4j-desktop"
export JAVA_HOME="$desktop/Application/Cache/runtime/zulu21.44.17-ca-jdk21.0.8-macosx_aarch64/"
$desktop/Application/Data/dbmss/dbms-04e69e6f-cb15-42a6-a26e-908bc7d09d77bin/neo4j-admin \
        database import incremental pep \
        --force --verbose \
        --id-type=STRING \
        --relationships=_links.csv \
        --nodes=_nodes.csv \
        --nodes=RealEstate.csv \
        --relationships=Ownership.csv \
        --nodes=Vehicle.csv \
        --nodes=Person.csv \
        --nodes=LegalEntity.csv \
        --relationships=Payment.csv \

import errors

the import starts ok but fails immediately with this error: No label was specified for the node index in 'id:ID'

also seems to be related to a null str (Cannot invoke "String.length()" because "str" is null) but i cant find any nulls in the csv, just blank strings

doesnt matter if i do full or incremental, same error

example node

"id:ID",":LABEL","caption","name","amount","address","alias","amountEur","amountUsd","country","createdAt","currency","description","keywords","modifiedAt","notes","previousName","program","publisher","publisherUrl","retrievedAt","sourceUrl","summary","topics","weakAlias","wikidataId","wikipediaUrl"
"entity:000044504e4b03630baf82dc4114f4c16f613e3d.d0833ede83bff05a9d59d559d70f801bc83c5d82","Thing;Asset;Value","something","asdfasdf","10000.0","","","","","","","","123 fake st","","","","","","","","","","","","","",""

notice the label value is ; sep but i think it should be : sep. even after fixing i get the same error

full trace attached neo4j-admin-exception-trace-2026-03-11.06.13.51.log

[pudo]

Hey @c4jquick! I haven't really dug into the bulk neo4j exporter for a while, the workflow had been a bit cursed because the server always needs to be able to resolve the paths in the import scripts and that makes it not very transferable. Do you see benefits to this approach over the direct DB connection idea sketched here: https://github.com/opensanctions/followthemoney-graph ? My sense is that this, with the right batching, isn't very different in terms of performance.

[c4jquick]

so the neo4j csv import using the built in util neo4j-admin should be a lot faster because it runs on the local dbms filesystem and can even import while instance is stopped. in doing a bolt/neo4j connection theres lots of overhead and asyncio wrangling to try and cut down the overhead but theres always the tcp layer too. the cypher exports work ok but very slowly (im trying to shove 10G of cypher). we already use ftm a lot in our data pipelines and store the emitted jsonl outputs from ftm map in s3. trying to use that as our starting point without having to know the original mapping file or details about the schema since we are doing this repeatedly on hundreds of datasets w/ individual schemata. ftm graph seems to combine both which will be nice to have some more advanced controls esp over relationships but w/ our current pipeline we just want a straightforward, atomic step of getting s3 files into neo4j quickly

[pudo]

yeah that's super fair. and if you have a standardization towards s3, that create a natural nexus. Just to share the slightly chaotic history here: I re-wrote the CSV dumper in ftm main to https://github.com/opensanctions/offshore-graph and then wanted to make it configurable via the -graph thing. I still think the mappings there could be fun to have. Perhaps CSV could be a sort of backend to the entity data generated by that thing?

btw the index error doesn't ring a bell for me immediately, but your analysis that the label spec is invalid seems super plausible!

[c4jquick]

running into a neo4j warning after about 400k nodes inserted using cypher because of the of the MATCH -> CREATE rel.

The disconnected pattern '(s: {...}), (t: {i...})' builds a cartesian product. A cartesian product may produce a large amount of data and slow down query processing.

its causing a big slow down from 1k/s to 400/s and only getting slower. apparently thats the best way to create a rel anyway so i couldnt found a way around that. another good reason why csvs would be ideal.

[c4jquick]

found the issue because neo4j apparently now has a much different csv serialization format. can combine nodes/edges in one file and uses apoc which should be more performant anyway. ill make a pr to update the exports and neo4j-admin calls to match the newest version

https://neo4j.com/labs/apoc/4.1/export/csv/

"_id","_labels","born","name","released","tagline","title","_start","_end","_type","roles"
"189",":Person","1964","Keanu Reeves","","","",,,,
"190",":Person","1967","Carrie-Anne Moss","","","",,,,
"191",":Person","1961","Laurence Fishburne","","","",,,,
"192",":Person","1960","Hugo Weaving","","","",,,,
"188",":Movie","","","1999","Welcome to the Real World","The Matrix",,,,
,,,,,,,"189","188","ACTED_IN","[""Neo""]"
,,,,,,,"190","188","ACTED_IN","[""Trinity""]"
,,,,,,,"191","188","ACTED_IN","[""Morpheus""]"
,,,,,,,"192","188","ACTED_IN","[""Agent Smith""]"