Commit 86b1a802 authored by Ramon Nou's avatar Ramon Nou
Browse files

test(malleability): isolate mutate hostfiles in tests 

Use separate mutate hostfiles for expand workflows instead of rewriting the
active hostfile, and remove redundant marker rewrites from error-handling
tests. This keeps daemon discovery state intact while still exercising
malleability commands with the intended hostfile mutations.
parent 6d690721

tests/integration/malleability/test_malleability_error_handling.py

+0 −31
Original line number Diff line number Diff line
@@ -67,17 +67,6 @@ def test_expand_start_with_same_node_count(gkfwd_daemon_factory, gkfs_shell): 
            if not line.startswith("#"):

                hf_out.write(line)



    # Use marker-based approach: no --new-hosts-file needed

    # Mark all entries with '+' to simulate adding same nodes (will fail gracefully)

    with open(hostfile, 'r') as hf_in:

        original_lines = hf_in.readlines()

    with open(hostfile, 'w') as hf_out:

        for line in original_lines:

            if not line.startswith('#') and line.strip():

                hf_out.write('+' + line)

            else:

                hf_out.write(line)



    cmd_str = f"LIBGKFS_HOSTS_FILE={hostfile} gkfs_malleability mutate start"

    cmd = gkfs_shell.script(cmd_str, intercept_shell=False, timeout=340)

    time.sleep(3)
@@ -105,16 +94,6 @@ def test_shrink_status_after_failed_expand(gkfwd_daemon_factory, gkfs_shell): 
            if not line.startswith("#"):

                hf_out.write(line)



    # Try mutate with markers (may fail due to same node count)

    with open(hostfile, 'r') as hf_in:

        original_lines = hf_in.readlines()

    with open(hostfile, 'w') as hf_out:

        for line in original_lines:

            if not line.startswith('#') and line.strip():

                hf_out.write('+' + line)

            else:

                hf_out.write(line)



    cmd_str = f"LIBGKFS_HOSTS_FILE={hostfile} gkfs_malleability mutate start"

    cmd = gkfs_shell.script(cmd_str, intercept_shell=False, timeout=340)

    time.sleep(3)
@@ -154,16 +133,6 @@ def test_malleability_expand_with_data(gkfwd_daemon_factory, gkfs_client, gkfs_s 
    cmd = gkfs_shell.script(cmd_str, intercept_shell=False)

    assert cmd.exit_code == 0



    # Start mutation with markers (fails due to same node count, which is expected)

    with open(hostfile, 'r') as hf_in:

        original_lines = hf_in.readlines()

    with open(hostfile, 'w') as hf_out:

        for line in original_lines:

            if not line.startswith('#') and line.strip():

                hf_out.write('+' + line)

            else:

                hf_out.write(line)



    cmd_str = f"LIBGKFS_HOSTS_FILE={hostfile} gkfs_malleability mutate start"

    cmd = gkfs_shell.script(cmd_str, intercept_shell=False, timeout=340)

tests/integration/malleability/test_malleability_tool_simple.py

+14 −12
Original line number Diff line number Diff line
@@ -81,9 +81,10 @@ def test_expand(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 
    d01 = gkfwd_daemon_factory.create(expand_mode=True)

    time.sleep(2)

    d01_port = d01.address.split(":")[-1]

    mutate_hostfile = hostfile.parent / "gkfs_hosts_expand_mutate.txt"

    with open(hostfile, "r") as hf_in:

        original_lines = hf_in.readlines()

    with open(hostfile, "w") as hf_out:

    with open(mutate_hostfile, "w") as hf_out:

        for line in original_lines:

            stripped = line.strip()

            if not stripped or stripped.startswith("#"):
@@ -94,23 +95,23 @@ def test_expand(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 
            else:

                hf_out.write(line)



    cmd = _run_cmd(hostfile, "mutate status", gkfs_shell)

    cmd = _run_cmd(mutate_hostfile, "mutate status", gkfs_shell)

    assert "No mutate running/finished." in cmd.stderr.decode()



    cmd = _run_cmd(hostfile, "mutate start", gkfs_shell, timeout=340

    cmd = _run_cmd(mutate_hostfile, "mutate start", gkfs_shell, timeout=340

    )

    assert "Mutate process" in cmd.stderr.decode()



    deadline = time.time() + 120

    while time.time() < deadline:

        cmd = _run_cmd(hostfile, "mutate status", gkfs_shell)

        cmd = _run_cmd(mutate_hostfile, "mutate status", gkfs_shell)

        if "No mutate running/finished." in cmd.stderr.decode():

            break

        time.sleep(2)

    else:

        pytest.fail("Mutate redistribution did not finish within 120 s")



    cmd = _run_cmd(hostfile, "mutate finalize", gkfs_shell)

    cmd = _run_cmd(mutate_hostfile, "mutate finalize", gkfs_shell)

    assert cmd.exit_code == 0



    for i in range(4):
@@ -118,7 +119,7 @@ def test_expand(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 
        ret = gkfs_client.stat(fpath)

        assert ret.retval == 0, f"stat failed for {fpath} after expand"



    with open(hostfile, "r") as hf:

    with open(mutate_hostfile, "r") as hf:

        content = hf.read()

    for line in content.splitlines():

        stripped = line.strip()
@@ -229,9 +230,10 @@ def test_mutate_swap(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 


    d01_port = d01.address.split(":")[-1]

    d02_port = d02.address.split(":")[-1]

    mutate_hostfile = old_hostfile.parent / "gkfs_hosts_mutate_swap.txt"

    with open(old_hostfile, "r") as hf_in:

        original_lines = hf_in.readlines()

    with open(old_hostfile, "w") as hf_out:

    with open(mutate_hostfile, "w") as hf_out:

        for line in original_lines:

            stripped = line.strip()

            if not stripped or stripped.startswith("#"):
@@ -244,23 +246,23 @@ def test_mutate_swap(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 
            else:

                hf_out.write(line)



    cmd = _run_cmd(old_hostfile, "mutate status", gkfs_shell)

    cmd = _run_cmd(mutate_hostfile, "mutate status", gkfs_shell)

    assert "No mutate running/finished." in cmd.stderr.decode()



    cmd = _run_cmd(old_hostfile, "mutate start", gkfs_shell, timeout=340

    cmd = _run_cmd(mutate_hostfile, "mutate start", gkfs_shell, timeout=340

    )

    assert "Mutate process" in cmd.stderr.decode()



    deadline = time.time() + 120

    while time.time() < deadline:

        cmd = _run_cmd(old_hostfile, "mutate status", gkfs_shell)

        cmd = _run_cmd(mutate_hostfile, "mutate status", gkfs_shell)

        if "No mutate running/finished." in cmd.stderr.decode():

            break

        time.sleep(2)

    else:

        pytest.fail("Mutate redistribution did not finish within 120 s")



    cmd = _run_cmd(old_hostfile, "mutate finalize", gkfs_shell)

    cmd = _run_cmd(mutate_hostfile, "mutate finalize", gkfs_shell)

    assert cmd.exit_code == 0



    for i in range(4):
@@ -268,7 +270,7 @@ def test_mutate_swap(gkfwd_daemon_factory, gkfs_client, gkfs_shell): 
        ret = gkfs_client.stat(fpath)

        assert ret.retval == 0, f"stat failed for {fpath} after mutate swap"



    with open(old_hostfile, "r") as hf:

    with open(mutate_hostfile, "r") as hf:

        content = hf.read()

    for line in content.splitlines():

        stripped = line.strip()