Why queries are slow during load?

Users
1

while i am running my code inserting data to dgraph, then even simple queries run very slow, like ddd(func: has(type_Work)) { count(uid) } query takes 40 seconds for 200k edges only
when i stop loading data, queries run normally.
so, in highly updated systems, there is no way to have realtime queries, am i right?
if it’s because of transactions (edges are locked for read, whatever), is it a way to disable transactions since we don’t use them at all?

2

nops, as far know, transactions are intrinsic to Dgraph at this point. And I do not imagine that’s the problem.

You can implement Apollo Graphql and use subscriptions to this.

The staff is working on improvements in the Has function.

Cheers.

3

it’s not about “has” function, when load stops, query runs for milliseconds even on millions of edges. but when i start loading data, it takes up to tens of minutes.

that’s the problem

4

btw, i have a lot of “old” transactions in logs. may be it’s the case:

2018/10/19 18:17:04 draft.go:769: Found 117 old transactions. Acting to abort them.
2018/10/19 18:17:04 draft.go:746: TryAbort 117 txns with start ts. Error: <nil>
2018/10/19 18:19:05 draft.go:772: abortOldTransactions for 117 txns. Error: <nil>
2018/10/19 18:20:08 draft.go:769: Found 192 old transactions. Acting to abort them.
2018/10/19 18:20:08 draft.go:746: TryAbort 192 txns with start ts. Error: <nil>
2018/10/19 18:23:48 draft.go:769: Found 64 old transactions. Acting to abort them.
2018/10/19 18:23:48 draft.go:746: TryAbort 64 txns with start ts. Error: <nil>
2018/10/19 18:24:02 draft.go:772: abortOldTransactions for 192 txns. Error: <nil>
2018/10/19 18:24:42 draft.go:772: abortOldTransactions for 64 txns. Error: <nil>
2018/10/19 18:26:03 draft.go:769: Found 84 old transactions. Acting to abort them.
2018/10/19 18:26:04 draft.go:746: TryAbort 84 txns with start ts. Error: <nil>
2018/10/19 18:27:17 draft.go:772: abortOldTransactions for 84 txns. Error: <nil>
2018/10/19 18:29:07 draft.go:769: Found 11 old transactions. Acting to abort them.
2018/10/19 18:29:07 draft.go:746: TryAbort 11 txns with start ts. Error: <nil>
2018/10/19 18:30:51 draft.go:772: abortOldTransactions for 11 txns. Error: <nil>
2018/10/19 18:32:44 draft.go:769: Found 227 old transactions. Acting to abort them.
2018/10/19 18:32:44 draft.go:746: TryAbort 227 txns with start ts. Error: <nil>

but in data load procedure i use autocommit, so have no idea how they appear:

  def runMutation(q: String): Unit = {
    val txn = dgraphClient.newTransaction
    try {
      val mu = Mutation.newBuilder()
        .setSetNquads(ByteString.copyFromUtf8(q))
        .setCommitNow(true)
        .setIgnoreIndexConflict(true)
        .build
      txn.mutate(mu)
    } catch {
      case e: Exception => {
        println(e.getMessage)
        //println(q)
      }
    } finally {
      txn.discard()
    }
  }
5

nope, i turned off parallel load, “old” transactions errors disappeared (now have just few “A tick missed to fire. Node blocks too long” ones), but simple “has” query still takes a lot of time when load procedure is running

6

What Dgraph version do you use?
Please, do this https://docs.dgraph.io/v1.0.9/deploy/#upgrade-database and upgrade if so.

7

i am using the last one, 1.0.9

8

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.