Description
What version of Dgraph are you using?
v1.1.1, v1.2.2, v20.03.0
Have you tried reproducing the issue with the latest release?
No. Latest is currently v20.03.1.
What is the hardware spec (RAM, OS)?
Two different machines running Ubuntu 16.04, one with 4cpu/30gb ram, the other with 8cpu/64gb ram.
Steps to reproduce the issue (command/config used to run Dgraph).
dgraph bulk -f {directory with rdf files} -s {schema file} --map_shards=2 --reduce_shards=1 --http=localhost:8000 --zero=localhost:5080 --format=rdf
Expected behaviour and actual result.
On both aforementioned machines, I tried running v1.2.2 and v20.03.0 with the same result: after successfully completing the MAP phase, the reduce phase failed with ~98.5M edge count (note that this edge count was consistent in the failures over both versions and both machines). Over a short period of time, the bulk loader would ramp up its memory usage to the entirety of the machine's RAM, then the bulk loader would freeze and crash.
When I downgraded the dgraph version to v1.1.1, the MAP and REDUCE phases completely successfully, and I was able to run a new dgraph cluster on top of the resultant p directory.