memory leak？ansys died and python become unresponsive

[war16641]

pymapdl is used to handle data at postprocess stage. the purpose of my script is to get the max stress at different element sets in all the loadsteps. the code is below. the problem is , i have many loadsteps and many element sets. with the script going, the memory used by ANSYS increase rapidly, the memory used by python script increase slightly. when the memory used by ANSYS reach to 15000MB, ansys died and python become unresponsive.
platform info: pycharm, python3.12, ansys 2024r1, pyansys 2025.1.2, windows 11

stress_out_dic = {}
for timei in range(1, numstep + 1): # take loadstep as iterator. here,numstep is 17. if set it to 12, this script will run normally.
    print(f'工况{timei}/{numstep}')
    mapdl.allsel()
    mapdl.set(time=timei)
    for u in tqdm.tqdm(fdm): # take element set(ansys compnent) as iterator.
        jdnm = u['节点']
        mapdl.cmsel('s','el_jd_' + jdnm)
        mapdl.nsle()
        jdstress = stress_out_dic.get(jdnm)
        if jdstress is None:
            numnd = mc.selednum('node')
            stress_out_dic[jdnm] = {
                's1max': np.zeros(numnd),
                's3min': np.zeros(numnd),
                'seqvmax': np.zeros(numnd),
                'max_step': 0,
            }
            jdstress = stress_out_dic[jdnm]

        # ANSYS died at the these 3 lines below.
        s1 = mapdl.post_processing.nodal_principal_stress('1')
        s3 = mapdl.post_processing.nodal_principal_stress('3')
        seqv = mapdl.post_processing.nodal_eqv_stress()
        # 比较取不利
        if np.max(seqv)> np.max(jdstress['seqvmax']):
            jdstress['max_step']=timei
        jdstress['s1max'] = np.maximum(jdstress['s1max'], s1)
        jdstress['s3min'] = np.minimum(jdstress['s3min'], s3)
        jdstress['seqvmax'] = np.maximum(jdstress['seqvmax'], seqv)

[SemTiOne]

I think that's not a Python leak, the growth is in the ANSYS process. nodal_principal_stress() / nodal_eqv_stress() both pull a full-model array every call and only mask down to your selection afterward in Python, so cmsel/nsle don't reduce what ANSYS computes. You're re-fetching the whole model once per element set instead of once per load step.

I suggest pulling the three arrays once per load step, then slicing per element set locally (also swapped mc.selednum('node') for len(sel_nodes) since I couldn't find that as a pymapdl API):

node_numbers_all = mapdl.mesh.nnum_all
node_to_idx = {n: i for i, n in enumerate(node_numbers_all)}
stress_out_dic = {}
for timei in range(1, numstep + 1):
    print(f'工况{timei}/{numstep}')
    mapdl.allsel()
    mapdl.set(time=timei)
    s1_all = mapdl.post_processing.nodal_principal_stress('1')
    s3_all = mapdl.post_processing.nodal_principal_stress('3')
    seqv_all = mapdl.post_processing.nodal_eqv_stress()
    for u in tqdm.tqdm(fdm):
        jdnm = u['节点']
        mapdl.cmsel('s', 'el_jd_' + jdnm)
        mapdl.nsle()
        sel_nodes = mapdl.mesh.nnum
        idx = np.array([node_to_idx[n] for n in sel_nodes])
        s1, s3, seqv = s1_all[idx], s3_all[idx], seqv_all[idx]
        jdstress = stress_out_dic.get(jdnm)
        if jdstress is None:
            numnd = len(sel_nodes)
            stress_out_dic[jdnm] = {
                's1max': np.zeros(numnd), 's3min': np.zeros(numnd),
                'seqvmax': np.zeros(numnd), 'max_step': 0,
            }
            jdstress = stress_out_dic[jdnm]
        if np.max(seqv) > np.max(jdstress['seqvmax']):
            jdstress['max_step'] = timei
        jdstress['s1max'] = np.maximum(jdstress['s1max'], s1)
        jdstress['s3min'] = np.minimum(jdstress['s3min'], s3)
        jdstress['seqvmax'] = np.maximum(jdstress['seqvmax'], seqv)

If it still grows, mapdl.result (reads the .rst file directly, DPF/reader-based) avoids touching the live session at all. Hope it helps!

[war16641]

just like you say, nodal_principal_stress() / nodal_eqv_stress() makes a memory copy of the whole model, regardless of the current selection of elements/nodes. that is the reason why ranning out of memory. but ,intuitively, nodal_principal_stress() / nodal_eqv_stress() just return the results of selection of elements/nodes, who knows it involve all the elements/nodes.
based on your code, i improve the line "node_numbers_all = mapdl.mesh.nnum_all". this line takes a very long time. finally, my script can be finished within 2 minutes and ansys memory occupation under 6000MB. thank you for the answer.

    # <editor-fold desc="节点成组单元提取idx 为nodal_principal_stress做准备">
    stime1=time.time()
    mapdl.esel('none')
    for u in fdm:
        jdnm = u['节点']
        mapdl.cmsel('a','el_jd_' + jdnm)
    mapdl.nsle()
    mapdl.cm('el_jd_allsx','ELEM')#所有会用到的单元 即相关单元和相关节点
    sel_nodes = mapdl.mesh.nnum
    node_to_idx = {n: i for i, n in enumerate(sel_nodes)}#相关节点的idx
    print(f'相关节点idx耗时{time.time()-stime1}s')
    #再循环一次得的各个组的节点的idx
    stime1 = time.time()
    jd_node_idx_dic={}
    for u in fdm:
        jdnm = u['节点']
        mapdl.cmsel('s','el_jd_' + jdnm)
        mapdl.nsle()
        sel_nodes = mapdl.mesh.nnum
        jd_node_idx_dic[jdnm]=np.array([node_to_idx[n] for n in sel_nodes])
    print(f'各个组的节点的idx耗时{time.time() - stime1}s')
    # </editor-fold>

    # 开始上弦
    stress_out_dic = {}
    for timei in range(1, numstep + 1):
        print(f'工况{timei}/{numstep}')
        mapdl.cmsel('s','el_jd_allsx')
        mapdl.nsle()
        mapdl.set(time=timei)
        s1_all = mapdl.post_processing.nodal_principal_stress('1')
        s3_all = mapdl.post_processing.nodal_principal_stress('3')
        seqv_all = mapdl.post_processing.nodal_eqv_stress()
        for u in tqdm.tqdm(fdm):
            jdnm = u['节点']
            idx=jd_node_idx_dic[jdnm]
            s1, s3, seqv = s1_all[idx], s3_all[idx], seqv_all[idx]
            jdstress = stress_out_dic.get(jdnm)
            if jdstress is None:
                numnd = len(idx)
                stress_out_dic[jdnm] = {
                    's1max': np.zeros(numnd), 's3min': np.zeros(numnd),
                    'seqvmax': np.zeros(numnd), 'max_step': 0,
                }
                jdstress = stress_out_dic[jdnm]
            if np.max(seqv) > np.max(jdstress['seqvmax']):
                jdstress['max_step'] = timei
            jdstress['s1max'] = np.maximum(jdstress['s1max'], s1)
            jdstress['s3min'] = np.minimum(jdstress['s3min'], s3)
            jdstress['seqvmax'] = np.maximum(jdstress['seqvmax'], seqv)