Possible division by zero in PageRank when graph contains zero vertices

[KHARSHAVARDHAN-eng]

Bug Type (问题类型)

others (please edit later)

Before submit

• 我已经确认现有的 Server-Issues 与 Computer-Issues 中没有相同 / 重复问题

Environment (环境信息)

Repository: apache/hugegraph-computer
Branch: master

Issue identified during source code review.
I have not yet verified it by executing the algorithm.

Expected & Actual behavior (期望与实际表现)

Expected Behavior

The PageRank implementation should safely handle graphs with zero vertices or explicitly validate this case before performing any division.

Observation

While reviewing the source code, I noticed a calculation that appears to divide by the total vertex count without an obvious check for a zero value.

If totalVertexCount can become zero, this could potentially result in a division-by-zero exception.

I have not yet been able to verify whether this execution path is reachable at runtime. I would appreciate confirmation from the maintainers on whether this scenario is possible or whether there is upstream validation that guarantees the value is always greater than zero.

If this is considered a valid issue, I would be happy to work on a fix and submit a pull request.

[dosubot]

Good catch — this does appear to be a valid concern. The division by totalVertexCount happens in PageRank.beforeSuperstep() in three places without any zero-check:

long totalVertex = context.totalVertexCount();
this.danglingRank = danglingTotalRank.value() / totalVertex;
this.initialRankInSuperstep = this.alpha / totalVertex;
this.initialValue = new DoubleValue(1.0 / totalVertex);

Looking at the upstream path, totalVertexCount() is a passthrough to superstepStat.vertexCount() [1] [2], and SuperstepStat simply accumulates counts from worker stats without enforcing any minimum [3]. There doesn't appear to be framework-level validation that guarantees the vertex count is always > 0 before PageRank's normalization runs.

Whether this is reachable at runtime depends on whether the framework would even schedule supersteps on an empty graph — the master's termination logic might stop iteration early if there are no active vertices — but there's no explicit guard, so a defensive check would be a reasonable improvement. A simple validation at the top of beforeSuperstep() (e.g., throwing an IllegalStateException if totalVertex == 0) would make this safe regardless of how the framework evolves.

To reply, just mention @dosu.

---

Share context across your team and agents. Try Dosu.

[imbajin]

Thanks for reporting this. We confirmed that the concern is valid. For an empty graph, the input SuperstepStat is still active by default, so PageRank can enter superstep 0 with totalVertexCount == 0 and perform these floating-point divisions. In Java this produces non-finite values (Infinity/NaN) rather than an ArithmeticException, but the lack of explicit empty-graph handling is still a robustness issue.

A reasonable fix with a focused test is welcome. Please make sure the empty-graph behavior is handled explicitly and does not affect the normal PageRank path.

As a heads-up, our current development and maintenance focus is mainly on Vermeer, the in-memory graph computing engine in this repository. If you are interested in contributing further, Vermeer is the area we especially encourage you to explore and focus on.