In systems such as these, it's very, very easy to end up with a circular chain of constraints (think of a snake biting its own tail). Now physically in the real system that doesn't matter - but mathematically the 'closing constraint' in the circuit is redundant (or at least partially so - it may be partly required to set a position or orientation). In addition to the mathematical redundancy, the constraint solver doesn't have an obvious starting point in the chain to work from.
Using less restrictive constraints in some part of the circuit can help - maybe constrain to a point, rather than an axis OR to an edge/axis rather than a flat face. When constraining items together, it's all too easy to inadvertently duplicate an existing aspect of previous constraints, leading to 'over-constrained'.
For the case shown by Stephen coaxial constraint to one crankpin implies that connecting rod will be parallel to the wheels - adding a similar constraint to another crankpin duplicates that implied parallelism and will give 'over constrained'. Constraining a point in the connecting rod bore, with the axis of a crank pin would not duplicate the implied wheel to con-rod parallelism. Some of this will depend upon exactly how other parts/sub-assemblies were constrained.
These issue can be very tricky to spot the cause of, so I always advise that as soon as any constraint fails investigate immediately. Going back one step, or re-specifying the last constraint done is reasonably straightforward, if things are left until more constraints are added it rapidly becomes almost impossible to diagnose the issue.