Merge pull request #3641 from CliMA/zs/ci_config

set use_krylov_method to false in the flame graph job

Author: szy21

config/perf_configs/flame_perf_target_diffusion.yml

@@ -7,9 +7,6 @@ dt_save_to_sol: "Inf"
 log_progress: false
 implicit_diffusion: true
 approximate_linear_solve_iters: 2
-use_krylov_method: true
-use_dynamic_krylov_rtol: false
-krylov_rtol: 0.99
 moist: "equil"
 surface_setup: "DefaultMoninObukhov"
 rad: "allskywithclear"

perf/flame.jl

@@ -44,7 +44,7 @@ allocs_limit["flame_perf_target_tracers"] = 441_904
 allocs_limit["flame_perf_diagnostics"] = 10_677_144
 allocs_limit["flame_perf_target_diagnostic_edmfx"] = 11_183_232
 allocs_limit["flame_perf_target_prognostic_edmfx"] = 761_712
-allocs_limit["flame_perf_target_diffusion"] = 5_010_162_496
+allocs_limit["flame_perf_target_diffusion"] = 100_360
 allocs_limit["flame_perf_target_threaded"] = 2047_736
 allocs_limit["flame_perf_target_callbacks"] = 391_864
 allocs_limit["flame_perf_gw"] = 581_381_976
@@ -60,12 +60,6 @@ max_allocs_for_full_sampling = 10e6
 expected_allocs = get(allocs_limit, job_id, max_allocs_for_full_sampling)
 sampling_rate = expected_allocs <= max_allocs_for_full_sampling ? 1 : 0.01
 
-# Some jobs are problematic (the ones with Krylov mostly)
-# https://github.com/pfitzseb/ProfileCanvas.jl/issues/34
-if job_id in ("flame_perf_target_diffusion",)
-    sampling_rate = 0.001
-end
-
 # use new allocation profiler
 @info "collecting allocations with sampling rate $sampling_rate"
 Profile.Allocs.clear()