In Search of an Understandable Consensus Algorithm (Raft)
USENIX ATC, 2014 · paper · Computer science
Introduces Raft, a consensus algorithm for managing a replicated log, designed to be easier to understand than Paxos while equally efficient. A user study with 43 students found Raft easier to learn; it also adds a cluster membership change mechanism.
Checked: PDF from scs.stanford.edu, fetched 4 Oct 2026. Page numbers are those of that PDF.
2 findings
Reference or label Figure 2 caption: "The server behavior in the upper-left box is described as a set of rules that trigger independently and repeatedly". The upper-left box is 'State'; the rules are in the lower-right 'Rules for Servers' box.
- p. 4 Figure 2 caption
The server behavior in the upper-left box is described as a set of rules that trigger independently and repeatedly.
- p. 4 Figure 2 caption
The server behavior in the upper-left box is described as a set of rules
Judge's note: On the full p.4 image the boxes are: State at upper-left, RequestVote RPC at upper-right, AppendEntries RPC at lower-left, and Rules for Servers at lower-right. Only the lower-right box holds rule-style behaviour, so the caption points to the wrong box.
Second check: another model agreed
Report a problem with this finding
Reference or label Section 5.4.3 step 9: "entries that are committed indirectly, such as index 2 in Figure 8(d)". The caption puts the indirect commit of index 2 in panel (e), not (d).
- p. 9 Section 5.4.3, step 9
such as index 2 in Figure 8(d)
- p. 8 Figure 8 caption
as in (e), then this entry is committed (S5 cannot win an election). At this point all preceding entries in the log are committed as well.
Judge's note: The Figure 8 caption on p.8 says that in (d) S5 could be elected and overwrite the index-2 entry, and that only in (e), once S1 replicates its term-4 entry, 'all preceding entries in the log are committed as well'. In panel (d) of the figure nothing is committed at index 2. The step-9 pointer on p.9 therefore matches panel (e).
Second check: another model agreed
Report a problem with this finding
6 minor (typos and the like; most checked by the judge only)
Wording Section 8.2: "We have mechanically proven the Log Completeness Property using the TLA proof system [6]". The paper never defines a 'Log Completeness Property'; Figure 3 and Section 5.4 name 'Leader Completeness'.
- p. 12 Section 8.2
We have mechanically proven the Log Completeness Property us-ing the TLA proof system [6].
- p. 5 Figure 3
Leader Completeness: if a log entry is committed in a given term
- p. 12 Section 8.2
mechanically proven the Log Completeness Property
- p. 12 Section 8.2 Correctness
mechanically proven the Log Completeness Property us-
Only the authors can say what is right here.
Judge's note: A search of the text finds 'Log Completeness' only at p.12. Figure 3 and Sections 5.4, 5.4.3 and 5.6 all use 'Leader Completeness Property'. This is probably that property, but it could be a lemma from [28], so only the authors can say which property was mechanically proven.
Second check: another model agreed
Report a problem with this finding
Wording Figure 14 caption: "a broadcast time of roughly 15ms" is hard to reconcile with Section 8.3 presenting a 12–24ms election timeout as still within the timing requirement ("lowering the timeouts beyond this point violates Raft's timing requirement").
- p. 13 Figure 14 caption
cluster of five servers with a broadcast time of roughly 15ms.
- p. 13 Section 8.3
However, lowering the timeouts beyond this point violates Raft’s timing requirement
- p. 10 Section 5.6
The broadcast time should be an order of mag-nitude less than the election timeout
- p. 13 Figure 14 caption
The measurements were taken on a cluster of five servers with a broadcast time of roughly 15ms.
Only the authors can say what is right here.
Judge's note: Section 5.6 (p.9–10) requires broadcastTime ≪ electionTimeout, with the broadcast time about an order of magnitude below the election timeout. With a ~15ms broadcast time, a 12ms minimum timeout is already below the broadcast time. The heartbeat interval (half the minimum timeout, about 6ms) is also shorter than one broadcast. Either the 15ms figure or the claim about where the violation begins needs clarifying by the authors.
Second check: another model agreed
Report a problem with this finding
Typo Section 1: "improve understandability,including decomposition" has no space after the comma.
- p. 1 Section 1, paragraph 4
improve understandability,including decomposition
Judge's note: On the crop of p.1 the comma runs straight into 'including' with no visible gap. Other commas on the same line and nearby lines are followed by normal spaces, and the text layer also has no space here. Cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Section 9: "However, Raft has less mechanism that VR or ZooKeeper" should read 'than VR or ZooKeeper'.
- p. 13 Section 9, last paragraph
However, Raft has less mechanism that VR or ZooKeeper
- p. 13 Section 9, Related work
Raft has less mechanism that VR or ZooKeeper
Judge's note: The word 'that' is visible in the last paragraph of the right column on the p.13 image. The comparative needs 'than'.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [12]: "Commununications of the ACM 21, 7" should read 'Communications of the ACM'.
- p. 15 Reference [12]
Commununications of the ACM 21, 7
Judge's note: The misspelling is visible on the p.15 image. Reference [30] on the same page spells 'Communications of the ACM' correctly.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [18]: editor printed as "O. Baboaglu"; probably Babaoğlu (Babaoglu).
- p. 15 Reference [18]
O. Baboaglu
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the editor of Distributed Algorithms (Springer, with K. Marzullo) is Özalp Babaoğlu. The p.15 image does show 'Baboaglu', so the transposed letters are in the published PDF, not an extraction artefact.
Second check: none (judge only)