Error failed when calling List.rollup

Hi! I want to upload ~15GB gzipped rdf files To local Dgraph cluster which contain one zero and 3 alpha (all alpha located on same volume, and running in ludicrous mode). Everything run on one AWS instance r5a.24xlarge, 768 GB RAM, 2TB SSD, 96 cores. Dgraph version : v21.03.1

I am using dgraph live loader for this purpose, and I perform upload in several steps. I successfully uploaded ~5 GB of data, but after it I start to see in log of one alpha(s) a lot of error messages like that:

W0901 08:43:08.025933  343110 mvcc.go:148] Error failed when calling List.rollup: while encoding: cannot iterate through the list: cannot initialize iterator when calling List.iterate: cannot read initial list part for list with base key 00000000000000000000066c696e6b6564000000000000694b24: could not read list part with key 04000000000000000000066c696e6b6564000000000000694b240000000000000001: Key not found rolling up key [0 0 0 0 0 0 0 0 0 0 6 108 105 110 107 101 100 0 0 0 0 0 0 105 75 36]

And when I tried to perform some query, I also get similar error:

{
    q(func: type(Bucket)) @filter(eq(period, "hour")) @groupby(timestamp) {
  }
}' | jq
{
  "errors": [
    {
      "message": ": cannot retrieve UIDs from list with key 000000000000000000000b6467726170682e7479706502024275636b6574: cannot initialize iterator when calling List.iterate: cannot read initial list part for list with base key 000000000000000000000b6467726170682e7479706502024275636b6574: could not read list part with key 040000000000000000000b6467726170682e7479706502024275636b65740000000000000001: Key not found",
      "extensions": {
        "code": "ErrorInvalidRequest"
      }
    }
  ],
  "data": null
}

But this query work fine before. I continue to upload data, and when it passed ~13GB, live loader failed with error message:

[23:11:20Z] Elapsed: 05h22m40s Txns: 265151 N-Quads: 265151000 N-Quads/s [last 5s]: 16400 Aborts: 217
panic: rpc error: code = Unknown desc = cannot retrieve predicate information: No connection exists
goroutine 355 [running]:
github.com/dgraph-io/dgraph/dgraph/cmd/live.(*loader).upsertUids(0xc0004a7200, 0xc04269c000, 0x3e8, 0x3e8)
        /ext-go/1/src/github.com/dgraph-io/dgraph/dgraph/cmd/live/run.go:394 +0xeb3
github.com/dgraph-io/dgraph/dgraph/cmd/live.(*loader).processLoadFile.func1(0xc0004ea5a0, 0xc0004a7200, 0xc00078da10)
        /ext-go/1/src/github.com/dgraph-io/dgraph/dgraph/cmd/live/run.go:552 +0x56e
created by github.com/dgraph-io/dgraph/dgraph/cmd/live.(*loader).processLoadFile
        /ext-go/1/src/github.com/dgraph-io/dgraph/dgraph/cmd/live/run.go:487 +0xa5

Any ideas, why suddenly query stop working, and why live loader fail, on such powerful instance, and relative small dataset?

What is the timestamp predicate? Group by feature doesn’t work with value edges. Only with relational edges.

timestamp is datetime predicate indexed by hour. Actually I was able groupby on this predicate and count nodes, and on other non uid predicates also.

Has there been a resolution of this? I’m seeing it now too:
mvcc.go:148] Error failed when calling List.rollup: while encoding: