Invalid persistence of executed changesets with run-on-change/run-always

[fpavageau]

Liquigraph can only be run twice with the following changelog:

<changelog>
  <changeset id="create-node" run-always="true">
    <query>CREATE (n:Node)</query>
  </changeset>
</changelog>

The first time, it will record the execution with a MERGE query equivalent to:

MATCH (changelog:__LiquigraphChangelog)
CREATE (changelog)<-[:EXECUTED_WITHIN_CHANGELOG {order: 1}]-(:__LiquigraphChangeset {id: "create-node"})

(other properties of the changeset omitted). All is well.

The second time, due to the presence of the order property inside the MERGE clause, it will record another execution, despite the method being called "upsert":

MATCH (changelog:__LiquigraphChangelog)
CREATE (changelog)<-[:EXECUTED_WITHIN_CHANGELOG {order: 2}]-(:__LiquigraphChangeset {id: "create-node"})

If you try to run Liquibase a third time, it will complain that you have less declared changesets than have been executed (persisted).

I think the fix is actually simple. Instead of

MATCH (changelog:__LiquigraphChangelog)
MERGE (changelog)<-[:EXECUTED_WITHIN_CHANGELOG {order: {1}}]-(changeset:__LiquigraphChangeset {id: {2}})
ON MATCH SET  changeset.checksum = {3},
              changeset.query = {4}
ON CREATE SET changeset.author = {5},
              changeset.query = {4}
              changeset.checksum = {3}

the CHANGESET_UPSERT query should be

MATCH (changelog:__LiquigraphChangelog)
MERGE (changelog)<-[exec:EXECUTED_WITHIN_CHANGELOG]-(changeset:__LiquigraphChangeset {id: {2}})
ON MATCH SET  changeset.checksum = {3},
              changeset.query = {4}
ON CREATE SET changeset.author = {5},
              changeset.query = {4}
              changeset.checksum = {3},
              exec.order = {1}

(not tested at all, though, but it's the gist of the fix).

I'll try preparing a pull request.

[fbiville]

Cool, thx for the report (and the future fix)!

[fbiville]

I guess we can close this ;-)

[fpavageau]

Definitely, if I'd checked master before filing the issue, I would have seen it was already fixed.