NoSuchTable thrown from import_copy_of()

[finnschiermer]

This was originally reported in issue #3701, but it's a different error.

Copy of original report (by @ironage ):

Stack trace:

Starting test testCreateMultipleEmbeddedObjects
uncaught exception in notifier thread: N5realm11NoSuchTableE: No such table exists
Exception backtrace:
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(_ZNK5realm13ConstTableRef5checkEv+0x78) [0x7fc38d854188]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(_ZNK5realm13ConstTableRefptEv+0x9) [0x7fc38d8541d9]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(_ZN5realm11Transaction14import_copy_ofENS_13ConstTableRefE+0x19) [0x7fc38d7bf869]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(_ZN5realm5QueryC2EPKS0_PNS_11TransactionENS_13PayloadPolicyE+0x170) [0x7fc38d65b0c0]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(_ZN5realm11Transaction14import_copy_ofERNS_5QueryENS_13PayloadPolicyE+0x32) [0x7fc38d7bfde2]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x409db8) [0x7fc38d353db8]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x3f0542) [0x7fc38d33a542]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x404963) [0x7fc38d34e963]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x405275) [0x7fc38d34f275]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x41b16c) [0x7fc38d36516c]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0x41b210) [0x7fc38d365210]
/mnt/jenkins/workspace/realm_realm-js_PR-2901/compiled/node-v64_linux_x64/realm.node(+0xb91ce0) [0x7fc38dadbce0]
/lib64/libpthread.so.0(+0x7ea5) [0x7fc3909acea5]
/lib64/libc.so.6(clone+0x6d) [0x7fc3906d58dd]
terminate called after throwing an instance of 'realm::NoSuchTable'
  what():  No such table exists

Code:

testCreateMultipleEmbeddedObjects: function() {
        const realm = new Realm({schema: [schemas.HouseOwnerSchema, schemas.AddressSchema]});

        realm.write(() => {
            realm.create(schemas.HouseOwnerSchema.name, { name: "Ib", addresses: [
                { street: "Algade", city: "Nordby" },
                { street: "Skolevej", city: "Sydpynten" }
            ]});
            realm.create(schemas.HouseOwnerSchema.name, { name: "Petra", addresses: [
                { street: "Algade", city: "Nordby" }
            ]});
            realm.create(schemas.HouseOwnerSchema.name, { name: "Hans" });
        });

        let owners = realm.objects(schemas.HouseOwnerSchema.name).sorted("name");
        let addresses = realm.objects(schemas.AddressSchema.name).sorted("street");
        TestCase.assertEqual(owners.length, 3);
        TestCase.assertEqual(addresses.length, 3);

        const names = ["Hans", "Ib", "Petra"];
        for (let i = 0; i < names.length; i++) {
            TestCase.assertEqual(owners[i]["name"], names[i]);
        }

        let streets = ["Algade", "Algade", "Skolevej"];
        for (let i = 0; i < streets.length; i++) {
            TestCase.assertEqual(addresses[i]["street"], streets[i]);
        }

        realm.write(() => {
            addresses[0]["street"] = "Strandvejen";
        });

        streets = ["Algade", "Skolevej", "Strandvejen"];
        for (let i = 0; i < streets.length; i++) {
            TestCase.assertEqual(addresses[i]["street"], streets[i]);
        }

        realm.close();
    }

[finnschiermer]

The call stack indicates an attempt to "handover" a table from transaction A to B but at a point in time, where A is no longer valid.

[finnschiermer]

Since it happens on the notifier thread, one may speculate if perhaps high CPU load has delayed the thread in which we want to import to a point where the stuff we want to import is no longer valid. If so, this would point to a bug in either OS or JS violating the constraints for use of "import_copy_of".

[finnschiermer]

Cannot reproduce. Closing, presumed fixed by #3749.

[finnschiermer]

Possibly related to https://github.com/realm/realm-object-store/issues/1071

[tgoyne]

I think I may have spotted what could be causing this.

[finnschiermer]

Awesome! Waiting anxiously.

[tgoyne]

When a new ResultsNotifier is created, it copies the query from the source Results to the notifier Transaction on the worker thread without any locking involved. This is actually almost safe, but there's a TOCTOU race at https://github.com/realm/realm-core/blob/master/src/realm/object-store/impl/results_notifier.cpp#L219 and we'll hit this error if the source Realm is invalidated between the check and when we obtain the table key inside import_copy_of(). The timing required for this is very tight, which explains why it's a pretty rare crash.

The code currently is still very structured around both the idea that creating a SharedGroup is an expensive operation and having explicit handover objects. It seems pretty easy to fix this and simplify the code a little by adjusting it to better take advantage of how things work now.