LINKSET within transaction: fails to remove newly created records.

[foriam]

OrientDB Version: 3.0.9 edit: 3.0.10

Java Version: 8-191

OS: MacOS

Expected behavior

When adding and subsequently removing a record to a LINKSET within a transaction, it should not be there after commit.

Actual behavior

When adding and subsequently removing a record to a LINKSET within a transaction, it is there again after commit.

Steps to reproduce

See the following test written in groovy using the Java API:

/*
   Structure:

   WithLinks {
       links: [Linked]
   }
*/
OClass withLinks = db.createClass("WithLinks")
withLinks.createProperty("links", OType.LINKSET)
OClass linked = db.createClass("Linked")

/* A link must already be there */
OElement withLinks1 = db.newInstance("WithLinks")
OElement link1 = db.newInstance("Linked")
link1.save()
withLinks1.setProperty("links", [link1].toSet())
withLinks1.save()

/* Only in transaction - without transaction all OK */
db.begin()
    OElement withLinks1InTrans = db.load(withLinks1.identity)

    /* Add a new linked record */
    OElement link2 = db.newInstance("Linked")
    link2.save()
    withLinks1InTrans.getProperty("links").add(link2)
    withLinks1InTrans.save()

    /* Remove all from ORecordLazySet - if only link2 removed all OK */
    withLinks1InTrans.getProperty("links").remove(link1)
    withLinks1InTrans.getProperty("links").remove(link2)
    withLinks1InTrans.save()

    /* All seems OK before commit */
    assert withLinks1InTrans.getProperty("links").size() == 0
    assert withLinks1.getProperty("links").size() == 0
db.commit()

/* No difference if reload or not */
withLinks1 = withLinks1.reload()

/* Initial record was removed */
assert !withLinks1.getProperty("links").contains(link1)
/* Fails: why is link2 still in the set? */
assert !withLinks1.getProperty("links").contains(link2)

While the flow of adding and removing within one transaction is not something you'd intentionally do, it is impossible to prevent when using an object mapping layer and complex operations on persisted objects. Can lead to serious and in the worst case long unnoticed bugs.

[mmacfadden]

@foriam I think this might be related too or a duplicate of another issue I just posted yesterday. #8632

Except in my case, the link that is trying to be updated is not for a newly created item.

What do you think?

[foriam]

@mmacfadden Maybe! Critical differences, however:

• This issue does not exist if both elements of LINKSET are created either inside or outside the transaction.
• The above test runs through with a LINKMAP (as in execute() in transactions doesn't UPDATE records #8632), as well as with a LINKLIST - it seems an issue specific to LINKSET.
• This issue is exists for PLOCAL as well as MEMORY dbs.

So it would rather be an indirect relationship to the same root cause.

[mmacfadden]

@foriam Thanks for the insight. If the OrientDB team doesn't look into them in the next day or so I will probably take a stab and figuring both of these out.

[foriam]

Is there any update on this bug? Has anyone from the OrientDB team checked these issues and can comment how they could be addressed? Thanks!

[mmacfadden]

@luigidellaquila @tglman Any thoughts on this one as well as perhaps #8632?

[luigidellaquila]

Hi @mmacfadden @foriam

I'm working on it now, I'll let you know asap

Thanks

Luigi

[mmacfadden]

@luigidellaquila Thanks for the update. I am hopeful that the fix might also take care of #8632.

[mmacfadden]

Also, if you nee any help testing let me know.

[luigidellaquila]

Hi @mmacfadden @foriam

I pushed the fix, it will be released with v 3.0.12
I'm checking the other issue, but I doubt that the cause is the same...

Thanks

Luigi