Error when MOI tries to query the dual solution of a QCQP

[mtanneau]

I'm running into this error when solving a QCQP with CPLEX. (see below for link to instance).
It seems to be related to the absence/presence of dual variables.

Note that the error occurs within the call to MOI.optimize!; calling CPLEX.optimize!(cpx.inner) does not raise any error.

import CPLEX
import MathOptInterface
const MOI = MathOptInterface

mof = MOI.FileFormats.MOF.Model()
MOI.read_from_file(mof, "cgcp.mof")

cpx=  CPLEX.Optimizer()
MOI.copy_to(cpx, mof)

MOI.optimize!(cpx)

raises the error

Version identifier: 12.10.0.0 | 2019-11-26 | 843d4de
Tried aggregator 1 time.
QCP Presolve eliminated 1 rows and 23 columns.
Aggregator did 14 substitutions.
Reduced QCP has 30 rows, 48 columns, and 85 nonzeros.
Reduced QCP has 11 quadratic constraints.
Presolve time = 0.00 sec. (0.04 ticks)
Parallel mode: using up to 8 threads for barrier.
Number of nonzeros in lower triangle of A*A' = 368
Using Approximate Minimum Degree ordering
Total time for automatic ordering = 0.00 sec. (0.01 ticks)
Summary statistics for Cholesky factor:
  Threads                   = 8
  Rows in Factor            = 30
  Integer space required    = 47
  Total non-zeros in factor = 398
  Total FP ops to factor    = 7066
 Itn      Primal Obj        Dual Obj  Prim Inf Upper Inf  Dual Inf Inf Ratio
   0   5.5683686e+01   0.0000000e+00  2.57e+01  0.00e+00  5.18e+02  1.00e+00
   1   5.3021286e+01  -1.7264892e+00  2.57e+01  0.00e+00  5.18e+02  6.50e-01
   2   7.1670240e+00  -3.3115288e+01  2.55e+01  0.00e+00  5.14e+02  1.05e-01
   3  -6.8434212e+01  -8.0049405e+01  2.26e+01  0.00e+00  4.55e+02  5.75e-02
   4  -2.5242076e+01  -3.3640323e+01  1.31e+01  0.00e+00  2.65e+02  3.01e-01
   5  -6.1799726e+00  -8.1199681e+00  5.30e+00  0.00e+00  1.07e+02  2.78e+00
   6  -3.0068095e+00  -3.9322157e+00  1.04e+00  0.00e+00  2.10e+01  5.31e+00
   7  -1.3315820e+00  -1.7592574e+00  5.04e-01  0.00e+00  1.02e+01  7.54e+00
   8  -7.7572327e-01  -8.9679876e-01  2.54e-01  0.00e+00  5.12e+00  2.32e+01
   9  -4.3256118e-01  -4.5677784e-01  7.43e-02  0.00e+00  1.50e+00  1.09e+02
  10  -4.1508915e-01  -4.2372981e-01  1.51e-02  0.00e+00  3.05e-01  2.96e+02
  11  -3.9575998e-01  -3.9612075e-01  5.44e-03  0.00e+00  1.10e-01  7.05e+03
  12  -3.9503422e-01  -3.9509833e-01  2.27e-04  0.00e+00  4.59e-03  3.94e+04
  13  -3.9496659e-01  -3.9499011e-01  4.05e-05  0.00e+00  8.17e-04  1.03e+05
  14  -3.9491515e-01  -3.9492168e-01  1.50e-05  0.00e+00  3.03e-04  3.57e+05
  15  -3.9490664e-01  -3.9490784e-01  4.23e-06  0.00e+00  8.52e-05  1.88e+06
  16  -3.9490365e-01  -3.9490374e-01  7.81e-07  0.00e+00  1.58e-05  2.33e+07
CPLEX Error  1232: No dual solution exists.

ERROR: CPLEX.CplexError(1232, "CPLEX Error  1232: No dual solution exists.\n")
Stacktrace:
 [1] c_api_getdj at /home/mtanneau/.julia/packages/CPLEX/Hxpuk/src/cpx_solve.jl:150 [inlined]
 [2] optimize!(::CPLEX.Optimizer) at /home/mtanneau/.julia/packages/CPLEX/Hxpuk/src/MOI/MOI_wrapper.jl:1913
 [3] top-level scope at REPL[9]:1

(the first line is printed by CPLEX, the rest is Julia's error message).

There seems to be a disagreement between CPLEX and MOI over the dual solution.
MOI claims that the model is solved to optimality, with both primal and dual status at MOI.FEASIBLE_POINT.
CPLEX claims there is no dual solution.

Find the problematic instance here, or use the download link.

[odow]

Is this on the latest release? The status codes got updated.

[mtanneau]

Yes, latest release and on master. (Just tried master 40s ago)

[EDIT]: did you mean latest release of CPLEX? I'm using 12.10.0.0

[odow]

Current master gives

julia> MOI.get(cpx, MOI.TerminationStatus())
OPTIMAL::TerminationStatusCode = 1

julia> MOI.get(cpx, MOI.PrimalStatus())
FEASIBLE_POINT::ResultStatusCode = 1

julia> MOI.get(cpx, MOI.DualStatus())
NO_SOLUTION::ResultStatusCode = 0

But leaving this open until #295 is resolved.