def initialize_guests():
"""
Initialize guests (fill their memories with specified patterns).
"""
logging.info("Phase 1: filling guest memory pages")
for session in lsessions:
vm = lvms[lsessions.index(session)]
logging.debug("Turning off swap on vm %s", vm.name)
session.cmd("swapoff -a", timeout=300)
# Start the allocator
_start_allocator(vm, session, 60 * perf_ratio)
# Execute allocator on guests
for i in range(0, vmsc):
vm = lvms[i]
a_cmd = "mem = MemFill(%d, %s, %s)" % (ksm_size, skeys[i],
dkeys[i])
_execute_allocator(a_cmd, vm, lsessions[i], 60 * perf_ratio)
a_cmd = "mem.value_fill(%d)" % skeys[0]
_execute_allocator(a_cmd, vm, lsessions[i], 120 * perf_ratio)
# Let ksm_overcommit_guest.py do its job
# (until shared mem reaches expected value)
shm = 0
j = 0
logging.debug("Target shared meminfo for guest %s: %s", vm.name,
ksm_size)
while ((new_ksm and (shm < (ksm_size * (i + 1))))
or (not new_ksm and (shm < (ksm_size)))):
if j > 64:
logging.debug(utils_test.get_memory_info(lvms))
raise error.TestError("SHM didn't merge the memory until "
"the DL on guest: %s" % vm.name)
st = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", st)
time.sleep(st)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug(
"Shared meminfo for guest %s after "
"iteration %s: %s", vm.name, j, shm)
j += 1
# Keep some reserve
rt = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", rt)
time.sleep(rt)
logging.debug(utils_test.get_memory_info(lvms))
logging.info("Phase 1: PASS")
def initialize_guests():
"""
Initialize guests (fill their memories with specified patterns).
"""
logging.info("Phase 1: filling guest memory pages")
for session in lsessions:
vm = lvms[lsessions.index(session)]
logging.debug("Turning off swap on vm %s", vm.name)
session.cmd("swapoff -a", timeout=300)
# Start the allocator
_start_allocator(vm, session, 60 * perf_ratio)
# Execute allocator on guests
for i in range(0, vmsc):
vm = lvms[i]
cmd = "mem = MemFill(%d, %s, %s)" % (ksm_size, skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % skeys[0]
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)
# Let ksm_overcommit_guest.py do its job
# (until shared mem reaches expected value)
shm = 0
j = 0
logging.debug("Target shared meminfo for guest %s: %s", vm.name,
ksm_size)
while ((new_ksm and (shm < (ksm_size * (i + 1)))) or
(not new_ksm and (shm < (ksm_size)))):
if j > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until "
"the DL on guest: %s" % vm.name)
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo for guest %s after "
"iteration %s: %s", vm.name, j, shm)
j += 1
# Keep some reserve
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
logging.debug(utils_test.get_memory_info(lvms))
logging.info("Phase 1: PASS")
def separate_first_guest():
"""
Separate memory of the first guest by generating special random series
"""
logging.info("Phase 2: Split the pages on the first guest")
a_cmd = "mem.static_random_fill()"
data = _execute_allocator(a_cmd, lvms[0], lsessions[0],
120 * perf_ratio)[1]
r_msg = data.splitlines()[-1]
logging.debug("Return message of static_random_fill: %s", r_msg)
out = int(r_msg.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s", ksm_size,
out, (ksm_size * 1000 / out))
logging.debug(utils_test.get_memory_info(lvms))
logging.debug("Phase 2: PASS")
def separate_first_guest():
"""
Separate memory of the first guest by generating special random series
"""
logging.info("Phase 2: Split the pages on the first guest")
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, lvms[0], lsessions[0],
fill_base_timeout * 2 * perf_ratio)[1]
r_msg = data.splitlines()[-1]
logging.debug("Return message of static_random_fill: %s", r_msg)
out = int(r_msg.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s", ksm_size,
out, (ksm_size * 1000 / out))
logging.debug(utils_test.get_memory_info(lvms))
logging.debug("Phase 2: PASS")
def split_parallel():
"""
Parallel page spliting
"""
logging.info("Phase 1: parallel page spliting")
# We have to wait until allocator is finished (it waits 5 seconds to
# clean the socket
session = lsessions[0]
vm = lvms[0]
for i in range(1, max_alloc):
lsessions.append(vm.wait_for_login(timeout=360))
session.cmd("swapoff -a", timeout=300)
for i in range(0, max_alloc):
# Start the allocator
_start_allocator(vm, lsessions[i], 60 * perf_ratio)
logging.info("Phase 1: PASS")
logging.info("Phase 2a: Simultaneous merging")
logging.debug("Memory used by allocator on guests = %dMB",
(ksm_size / max_alloc))
for i in range(0, max_alloc):
a_cmd = "mem = MemFill(%d, %s, %s)" % (
(ksm_size / max_alloc), skeys[i], dkeys[i])
_execute_allocator(a_cmd, vm, lsessions[i], 60 * perf_ratio)
a_cmd = "mem.value_fill(%d)" % (skeys[0])
_execute_allocator(a_cmd, vm, lsessions[i], 90 * perf_ratio)
# Wait until ksm_overcommit_guest.py merges the pages (3 * ksm_size / 3)
shm = 0
i = 0
logging.debug("Target shared memory size: %s", ksm_size)
while (shm < ksm_size):
if i > 64:
logging.debug(utils_test.get_memory_info(lvms))
raise error.TestError("SHM didn't merge the memory until DL")
wt = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceed...", wt)
time.sleep(wt)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo after attempt %s: %s", i, shm)
i += 1
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2a: PASS")
logging.info("Phase 2b: Simultaneous spliting")
# Actual splitting
for i in range(0, max_alloc):
a_cmd = "mem.static_random_fill()"
data = _execute_allocator(a_cmd, vm, lsessions[i],
90 * perf_ratio)[1]
data = data.splitlines()[-1]
logging.debug(data)
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
(ksm_size / max_alloc), out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2b: PASS")
logging.info("Phase 2c: Simultaneous verification")
for i in range(0, max_alloc):
a_cmd = "mem.static_random_verify()"
data = _execute_allocator(a_cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2c: PASS")
logging.info("Phase 2d: Simultaneous merging")
# Actual splitting
for i in range(0, max_alloc):
a_cmd = "mem.value_fill(%d)" % skeys[0]
data = _execute_allocator(a_cmd, vm, lsessions[i],
120 * perf_ratio)[1]
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2d: PASS")
logging.info("Phase 2e: Simultaneous verification")
for i in range(0, max_alloc):
a_cmd = "mem.value_check(%d)" % skeys[0]
data = _execute_allocator(a_cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2e: PASS")
logging.info("Phase 2f: Simultaneous spliting last 96B")
for i in range(0, max_alloc):
a_cmd = "mem.static_random_fill(96)"
data = _execute_allocator(a_cmd, vm, lsessions[i],
60 * perf_ratio)[1]
data = data.splitlines()[-1]
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
ksm_size / max_alloc, out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2f: PASS")
logging.info("Phase 2g: Simultaneous verification last 96B")
for i in range(0, max_alloc):
a_cmd = "mem.static_random_verify(96)"
_, data = _execute_allocator(a_cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2g: PASS")
logging.debug("Cleaning up...")
for i in range(0, max_alloc):
lsessions[i].cmd_output("die()", 20)
session.close()
vm.destroy(gracefully=False)
def split_guest():
"""
Sequential split of pages on guests up to memory limit
"""
logging.info("Phase 3a: Sequential split of pages on guests up to "
"memory limit")
last_vm = 0
session = None
vm = None
for i in range(1, vmsc):
# Check VMs
for j in range(0, vmsc):
if not lvms[j].is_alive:
e_msg = (
"VM %d died while executing static_random_fill in "
"VM %d on allocator loop" % (j, i))
raise error.TestFail(e_msg)
vm = lvms[i]
session = lsessions[i]
a_cmd = "mem.static_random_fill()"
logging.debug(
"Executing %s on ksm_overcommit_guest.py loop, "
"vm: %s", a_cmd, vm.name)
session.sendline(a_cmd)
out = ""
try:
logging.debug(
"Watching host memory while filling vm %s memory", vm.name)
while not out.startswith("PASS") and not out.startswith(
"FAIL"):
if not vm.is_alive():
e_msg = (
"VM %d died while executing static_random_fill"
" on allocator loop" % i)
raise error.TestFail(e_msg)
free_mem = int(utils.read_from_meminfo("MemFree"))
if (ksm_swap):
free_mem = (free_mem +
int(utils.read_from_meminfo("SwapFree")))
logging.debug("Free memory on host: %d", free_mem)
# We need to keep some memory for python to run.
if (free_mem < 64000) or (ksm_swap and free_mem <
(450000 * perf_ratio)):
vm.pause()
for j in range(0, i):
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
logging.debug("Only %s free memory, killing %d guests",
free_mem, (i - 1))
last_vm = i
break
out = session.read_nonblocking(0.1, 1)
time.sleep(2)
except OSError:
logging.debug("Only %s host free memory, killing %d guests",
free_mem, (i - 1))
logging.debug("Stopping %s", vm.name)
vm.pause()
for j in range(0, i):
logging.debug("Destroying %s", lvms[j].name)
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
last_vm = i
if last_vm != 0:
break
logging.debug("Memory filled for guest %s", vm.name)
logging.info("Phase 3a: PASS")
logging.info("Phase 3b: Check if memory in max loading guest is right")
for i in range(last_vm + 1, vmsc):
lsessions[i].close()
if i == (vmsc - 1):
logging.debug(utils_test.get_memory_info([lvms[i]]))
logging.debug("Destroying guest %s", lvms[i].name)
lvms[i].destroy(gracefully=False)
# Verify last machine with randomly generated memory
a_cmd = "mem.static_random_verify()"
_execute_allocator(a_cmd, lvms[last_vm], lsessions[last_vm],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([lvms[last_vm]]))
lsessions[i].cmd_output("die()", 20)
lvms[last_vm].destroy(gracefully=False)
logging.info("Phase 3b: PASS")
if not lvms[i].is_alive():
raise error.TestError("VM %s seems to be dead; Test requires a"
"living VM" % lvms[i].name)
lsessions.append(lvms[i].wait_for_login(timeout=360))
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
raise error.TestFail("Could not get PID of %s" % (vm_name))
# Let guests rest a little bit :-)
st = vmsc * 2 * perf_ratio
logging.debug("Waiting %ds before proceed", st)
time.sleep(vmsc * 2 * perf_ratio)
logging.debug(utils_test.get_memory_info(lvms))
# Copy ksm_overcommit_guest.py into guests
virt_dir = os.path.join(os.environ['AUTODIR'], 'virt')
vksmd_src = os.path.join(virt_dir, "scripts", "ksm_overcommit_guest.py")
dst_dir = "/tmp"
for vm in lvms:
vm.copy_files_to(vksmd_src, dst_dir)
logging.info("Phase 0: PASS")
if params['ksm_mode'] == "parallel":
logging.info("Starting KSM test parallel mode")
split_parallel()
logging.info("KSM test parallel mode: PASS")
elif params['ksm_mode'] == "serial":
logging.info("Starting KSM test serial mode")
def split_parallel():
"""
Parallel page spliting
"""
logging.info("Phase 1: parallel page spliting")
# We have to wait until allocator is finished (it waits 5 seconds to
# clean the socket
session = lsessions[0]
vm = lvms[0]
for i in range(1, max_alloc):
lsessions.append(vm.wait_for_login(timeout=360))
session.cmd("swapoff -a", timeout=300)
for i in range(0, max_alloc):
# Start the allocator
_start_allocator(vm, lsessions[i], 60 * perf_ratio)
logging.info("Phase 1: PASS")
logging.info("Phase 2a: Simultaneous merging")
logging.debug("Memory used by allocator on guests = %dMB",
(ksm_size / max_alloc))
for i in range(0, max_alloc):
cmd = "mem = MemFill(%d, %s, %s)" % ((ksm_size / max_alloc),
skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % (skeys[0])
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)
# Wait until ksm_overcommit_guest.py merges pages (3 * ksm_size / 3)
shm = 0
i = 0
logging.debug("Target shared memory size: %s", ksm_size)
while (shm < ksm_size):
if i > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until DL")
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceed...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo after attempt %s: %s", i, shm)
i += 1
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2a: PASS")
logging.info("Phase 2b: Simultaneous spliting")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
logging.debug(data)
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
(ksm_size / max_alloc), out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2b: PASS")
logging.info("Phase 2c: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify()"
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2c: PASS")
logging.info("Phase 2d: Simultaneous merging")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.value_fill(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)[1]
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2d: PASS")
logging.info("Phase 2e: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.value_check(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2e: PASS")
logging.info("Phase 2f: Simultaneous spliting last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_fill(96)"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
ksm_size / max_alloc, out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2f: PASS")
logging.info("Phase 2g: Simultaneous verification last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify(96)"
_, data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2g: PASS")
logging.debug("Cleaning up...")
for i in range(0, max_alloc):
lsessions[i].cmd_output("die()", 20)
session.close()
vm.destroy(gracefully=False)
def split_guest():
"""
Sequential split of pages on guests up to memory limit
"""
logging.info("Phase 3a: Sequential split of pages on guests up to "
"memory limit")
last_vm = 0
session = None
vm = None
for i in range(1, vmsc):
# Check VMs
for j in range(0, vmsc):
if not lvms[j].is_alive:
e_msg = ("VM %d died while executing static_random_fill on"
" VM %d in allocator loop" % (j, i))
test.fail(e_msg)
vm = lvms[i]
session = lsessions[i]
cmd = "mem.static_random_fill()"
logging.debug("Executing %s on ksm_overcommit_guest.py loop, "
"vm: %s", cmd, vm.name)
session.sendline(cmd)
out = ""
try:
logging.debug("Watching host mem while filling vm %s memory",
vm.name)
while (not out.startswith("PASS") and
not out.startswith("FAIL")):
if not vm.is_alive():
e_msg = ("VM %d died while executing "
"static_random_fill on allocator loop" % i)
test.fail(e_msg)
free_mem = int(utils_memory.read_from_meminfo("MemFree"))
if (ksm_swap):
free_mem = (free_mem +
int(utils_memory.read_from_meminfo("SwapFree")))
logging.debug("Free memory on host: %d", free_mem)
# We need to keep some memory for python to run.
if (free_mem < 64000) or (ksm_swap and
free_mem < (450000 * perf_ratio)):
vm.pause()
for j in range(0, i):
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
logging.debug("Only %s free memory, killing %d guests",
free_mem, (i - 1))
last_vm = i
out = session.read_nonblocking(0.1, 1)
time.sleep(2)
except OSError:
logging.debug("Only %s host free memory, killing %d guests",
free_mem, (i - 1))
logging.debug("Stopping %s", vm.name)
vm.pause()
for j in range(0, i):
logging.debug("Destroying %s", lvms[j].name)
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
last_vm = i
if last_vm != 0:
break
logging.debug("Memory filled for guest %s", vm.name)
logging.info("Phase 3a: PASS")
logging.info("Phase 3b: Verify memory of the max stressed VM")
for i in range(last_vm + 1, vmsc):
lsessions[i].close()
if i == (vmsc - 1):
logging.debug(utils_test.get_memory_info([lvms[i]]))
logging.debug("Destroying guest %s", lvms[i].name)
lvms[i].destroy(gracefully=False)
# Verify last machine with randomly generated memory
cmd = "mem.static_random_verify()"
_execute_allocator(cmd, lvms[last_vm], lsessions[last_vm],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([lvms[last_vm]]))
lsessions[last_vm].cmd_output("die()", 20)
lvms[last_vm].destroy(gracefully=False)
logging.info("Phase 3b: PASS")
def run(test, params, env):
"""
Tests KSM (Kernel Shared Memory) capability by allocating and filling
KVM guests memory using various values. KVM sets the memory as
MADV_MERGEABLE so all VM's memory can be merged. The workers in
guest writes to tmpfs filesystem thus allocations are not limited
by process max memory, only by VM's memory. Two test modes are supported -
serial and parallel.
Serial mode - uses multiple VMs, allocates memory per guest and always
verifies the correct number of shared memory.
0) Prints out the setup and initialize guest(s)
1) Fills guest with the same number (S1)
2) Random fill on the first guest
3) Random fill of the remaining VMs one by one until the
memory is completely filled (KVM stops machines which
asks for additional memory until there is available
memory) (S2, shouldn't finish)
4) Destroy all VMs but the last one
5) Checks the last VMs memory for corruption
Parallel mode - uses one VM with multiple allocator workers. Executes
scenarios in parallel to put more stress on the KVM.
0) Prints out the setup and initialize guest(s)
1) Fills memory with the same number (S1)
2) Fills memory with random numbers (S2)
3) Verifies all pages
4) Fills memory with the same number (S2)
5) Changes the last 96B (S3)
Scenarios:
S1) Fill all vms with the same value (all pages should be merged into 1)
S2) Random fill (all pages should be splitted)
S3) Fill last 96B (change only last 96B of each page; some pages will be
merged; there was a bug with data corruption)
Every worker has unique random key so we are able to verify the filled
values.
:param test: kvm test object.
:param params: Dictionary with test parameters.
:param env: Dictionary with the test environment.
:param cfg: ksm_swap - use swap?
:param cfg: ksm_overcommit_ratio - memory overcommit (serial mode only)
:param cfg: ksm_parallel_ratio - number of workers (parallel mode only)
:param cfg: ksm_host_reserve - override memory reserve on host in MB
:param cfg: ksm_guest_reserve - override memory reserve on guests in MB
:param cfg: ksm_mode - test mode {serial, parallel}
:param cfg: ksm_perf_ratio - performance ratio, increase it when your
machine is too slow
"""
def _start_allocator(vm, session, timeout):
"""
Execute ksm_overcommit_guest.py on guest, wait until it's initialized.
:param vm: VM object.
:param session: Remote session to a VM object.
:param timeout: Timeout that will be used to verify if
ksm_overcommit_guest.py started properly.
"""
logging.debug("Starting ksm_overcommit_guest.py on guest %s", vm.name)
session.sendline("python /tmp/ksm_overcommit_guest.py")
try:
session.read_until_last_line_matches(["PASS:"******"FAIL:"], timeout)
except aexpect.ExpectProcessTerminatedError as details:
e_msg = ("Command ksm_overcommit_guest.py on vm '%s' failed: %s" %
(vm.name, str(details)))
test.fail(e_msg)
def _execute_allocator(command, vm, session, timeout):
"""
Execute a given command on ksm_overcommit_guest.py main loop,
indicating the vm the command was executed on.
:param command: Command that will be executed.
:param vm: VM object.
:param session: Remote session to VM object.
:param timeout: Timeout used to verify expected output.
:return: Tuple (match index, data)
"""
logging.debug("Executing '%s' on ksm_overcommit_guest.py loop, "
"vm: %s, timeout: %s", command, vm.name, timeout)
session.sendline(command)
try:
(match, data) = session.read_until_last_line_matches(
["PASS:"******"FAIL:"],
timeout)
except aexpect.ExpectProcessTerminatedError as details:
e_msg = ("Failed to execute command '%s' on "
"ksm_overcommit_guest.py, vm '%s': %s" %
(command, vm.name, str(details)))
test.fail(e_msg)
return (match, data)
def get_ksmstat():
"""
Return sharing memory by ksm in MB
:return: memory in MB
"""
fpages = open('/sys/kernel/mm/ksm/pages_sharing')
ksm_pages = int(fpages.read())
fpages.close()
return ((ksm_pages * 4096) / 1e6)
def initialize_guests():
"""
Initialize guests (fill their memories with specified patterns).
"""
logging.info("Phase 1: filling guest memory pages")
for session in lsessions:
vm = lvms[lsessions.index(session)]
logging.debug("Turning off swap on vm %s", vm.name)
session.cmd("swapoff -a", timeout=300)
# Start the allocator
_start_allocator(vm, session, 60 * perf_ratio)
# Execute allocator on guests
for i in range(0, vmsc):
vm = lvms[i]
cmd = "mem = MemFill(%d, %s, %s)" % (ksm_size, skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % skeys[0]
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)
# Let ksm_overcommit_guest.py do its job
# (until shared mem reaches expected value)
shm = 0
j = 0
logging.debug("Target shared meminfo for guest %s: %s", vm.name,
ksm_size)
while ((new_ksm and (shm < (ksm_size * (i + 1)))) or
(not new_ksm and (shm < (ksm_size)))):
if j > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until "
"the DL on guest: %s" % vm.name)
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo for guest %s after "
"iteration %s: %s", vm.name, j, shm)
j += 1
# Keep some reserve
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
logging.debug(utils_test.get_memory_info(lvms))
logging.info("Phase 1: PASS")
def separate_first_guest():
"""
Separate memory of the first guest by generating special random series
"""
logging.info("Phase 2: Split the pages on the first guest")
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, lvms[0], lsessions[0],
fill_base_timeout * 2 * perf_ratio)[1]
r_msg = data.splitlines()[-1]
logging.debug("Return message of static_random_fill: %s", r_msg)
out = int(r_msg.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s", ksm_size,
out, (ksm_size * 1000 / out))
logging.debug(utils_test.get_memory_info(lvms))
logging.debug("Phase 2: PASS")
def split_guest():
"""
Sequential split of pages on guests up to memory limit
"""
logging.info("Phase 3a: Sequential split of pages on guests up to "
"memory limit")
last_vm = 0
session = None
vm = None
for i in range(1, vmsc):
# Check VMs
for j in range(0, vmsc):
if not lvms[j].is_alive:
e_msg = ("VM %d died while executing static_random_fill on"
" VM %d in allocator loop" % (j, i))
test.fail(e_msg)
vm = lvms[i]
session = lsessions[i]
cmd = "mem.static_random_fill()"
logging.debug("Executing %s on ksm_overcommit_guest.py loop, "
"vm: %s", cmd, vm.name)
session.sendline(cmd)
out = ""
try:
logging.debug("Watching host mem while filling vm %s memory",
vm.name)
while (not out.startswith("PASS") and
not out.startswith("FAIL")):
if not vm.is_alive():
e_msg = ("VM %d died while executing "
"static_random_fill on allocator loop" % i)
test.fail(e_msg)
free_mem = int(utils_memory.read_from_meminfo("MemFree"))
if (ksm_swap):
free_mem = (free_mem +
int(utils_memory.read_from_meminfo("SwapFree")))
logging.debug("Free memory on host: %d", free_mem)
# We need to keep some memory for python to run.
if (free_mem < 64000) or (ksm_swap and
free_mem < (450000 * perf_ratio)):
vm.pause()
for j in range(0, i):
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
logging.debug("Only %s free memory, killing %d guests",
free_mem, (i - 1))
last_vm = i
out = session.read_nonblocking(0.1, 1)
time.sleep(2)
except OSError:
logging.debug("Only %s host free memory, killing %d guests",
free_mem, (i - 1))
logging.debug("Stopping %s", vm.name)
vm.pause()
for j in range(0, i):
logging.debug("Destroying %s", lvms[j].name)
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
last_vm = i
if last_vm != 0:
break
logging.debug("Memory filled for guest %s", vm.name)
logging.info("Phase 3a: PASS")
logging.info("Phase 3b: Verify memory of the max stressed VM")
for i in range(last_vm + 1, vmsc):
lsessions[i].close()
if i == (vmsc - 1):
logging.debug(utils_test.get_memory_info([lvms[i]]))
logging.debug("Destroying guest %s", lvms[i].name)
lvms[i].destroy(gracefully=False)
# Verify last machine with randomly generated memory
cmd = "mem.static_random_verify()"
_execute_allocator(cmd, lvms[last_vm], lsessions[last_vm],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([lvms[last_vm]]))
lsessions[last_vm].cmd_output("die()", 20)
lvms[last_vm].destroy(gracefully=False)
logging.info("Phase 3b: PASS")
def split_parallel():
"""
Parallel page spliting
"""
logging.info("Phase 1: parallel page spliting")
# We have to wait until allocator is finished (it waits 5 seconds to
# clean the socket
session = lsessions[0]
vm = lvms[0]
for i in range(1, max_alloc):
lsessions.append(vm.wait_for_login(timeout=360))
session.cmd("swapoff -a", timeout=300)
for i in range(0, max_alloc):
# Start the allocator
_start_allocator(vm, lsessions[i], 60 * perf_ratio)
logging.info("Phase 1: PASS")
logging.info("Phase 2a: Simultaneous merging")
logging.debug("Memory used by allocator on guests = %dMB",
(ksm_size / max_alloc))
for i in range(0, max_alloc):
cmd = "mem = MemFill(%d, %s, %s)" % ((ksm_size / max_alloc),
skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % (skeys[0])
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)
# Wait until ksm_overcommit_guest.py merges pages (3 * ksm_size / 3)
shm = 0
i = 0
logging.debug("Target shared memory size: %s", ksm_size)
while (shm < ksm_size):
if i > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until DL")
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceed...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo after attempt %s: %s", i, shm)
i += 1
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2a: PASS")
logging.info("Phase 2b: Simultaneous spliting")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
logging.debug(data)
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
(ksm_size / max_alloc), out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2b: PASS")
logging.info("Phase 2c: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify()"
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2c: PASS")
logging.info("Phase 2d: Simultaneous merging")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.value_fill(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)[1]
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2d: PASS")
logging.info("Phase 2e: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.value_check(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2e: PASS")
logging.info("Phase 2f: Simultaneous spliting last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_fill(96)"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
ksm_size / max_alloc, out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2f: PASS")
logging.info("Phase 2g: Simultaneous verification last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify(96)"
_, data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2g: PASS")
logging.debug("Cleaning up...")
for i in range(0, max_alloc):
lsessions[i].cmd_output("die()", 20)
session.close()
vm.destroy(gracefully=False)
# Main test code
logging.info("Starting phase 0: Initialization")
if process.run("ps -C ksmtuned", ignore_status=True).exit_status == 0:
logging.info("Killing ksmtuned...")
process.run("killall ksmtuned")
new_ksm = False
if (os.path.exists("/sys/kernel/mm/ksm/run")):
process.run("echo 50 > /sys/kernel/mm/ksm/sleep_millisecs", shell=True)
process.run("echo 5000 > /sys/kernel/mm/ksm/pages_to_scan", shell=True)
process.run("echo 1 > /sys/kernel/mm/ksm/run", shell=True)
e_up = "/sys/kernel/mm/transparent_hugepage/enabled"
e_rh = "/sys/kernel/mm/redhat_transparent_hugepage/enabled"
if os.path.exists(e_up):
process.run("echo 'never' > %s" % e_up, shell=True)
if os.path.exists(e_rh):
process.run("echo 'never' > %s" % e_rh, shell=True)
new_ksm = True
else:
try:
process.run("modprobe ksm")
process.run("ksmctl start 5000 100")
except process.CmdError as details:
test.fail("Failed to load KSM: %s" % details)
# host_reserve: mem reserve kept for the host system to run
host_reserve = int(params.get("ksm_host_reserve", -1))
if (host_reserve == -1):
try:
available = utils_memory.read_from_meminfo("MemAvailable")
except process.CmdError: # ancient kernels
utils_memory.drop_caches()
available = utils_memory.read_from_meminfo("MemFree")
# default host_reserve = UsedMem + one_minimal_guest(128MB)
# later we add 64MB per additional guest
host_reserve = ((utils_memory.memtotal() - available) / 1024 + 128)
# using default reserve
_host_reserve = True
else:
_host_reserve = False
# guest_reserve: mem reserve kept to avoid guest OS to kill processes
guest_reserve = int(params.get("ksm_guest_reserve", -1))
if (guest_reserve == -1):
# default guest_reserve = minimal_system_mem(256MB)
# later we add tmpfs overhead
guest_reserve = 256
# using default reserve
_guest_reserve = True
else:
_guest_reserve = False
max_vms = int(params.get("max_vms", 2))
overcommit = float(params.get("ksm_overcommit_ratio", 2.0))
max_alloc = int(params.get("ksm_parallel_ratio", 1))
# vmsc: count of all used VMs
vmsc = int(overcommit) + 1
vmsc = max(vmsc, max_vms)
if (params['ksm_mode'] == "serial"):
max_alloc = vmsc
if _host_reserve:
# First round of additional guest reserves
host_reserve += vmsc * 64
_host_reserve = vmsc
host_mem = (int(utils_memory.memtotal()) / 1024 - host_reserve)
ksm_swap = False
if params.get("ksm_swap") == "yes":
ksm_swap = True
# Performance ratio
perf_ratio = params.get("ksm_perf_ratio")
if perf_ratio:
perf_ratio = float(perf_ratio)
else:
perf_ratio = 1
if (params['ksm_mode'] == "parallel"):
vmsc = 1
overcommit = 1
mem = host_mem
# 32bit system adjustment
if "64" not in params.get("vm_arch_name"):
logging.debug("Probably i386 guest architecture, "
"max allocator mem = 2G")
# Guest can have more than 2G but
# kvm mem + 1MB (allocator itself) can't
if (host_mem > 3100):
mem = 3100
if os.popen("uname -i").readline().startswith("i386"):
logging.debug("Host is i386 architecture, max guest mem is 2G")
# Guest system with qemu overhead (64M) can't have more than 2G
if mem > 3100 - 64:
mem = 3100 - 64
else:
# mem: Memory of the guest systems. Maximum must be less than
# host's physical ram
mem = int(overcommit * host_mem / vmsc)
# 32bit system adjustment
if not params['image_name'].endswith("64"):
logging.debug("Probably i386 guest architecture, "
"max allocator mem = 2G")
# Guest can have more than 2G but
# kvm mem + 1MB (allocator itself) can't
if mem - guest_reserve - 1 > 3100:
vmsc = int(math.ceil((host_mem * overcommit) /
(3100 + guest_reserve)))
if _host_reserve:
host_reserve += (vmsc - _host_reserve) * 64
host_mem -= (vmsc - _host_reserve) * 64
_host_reserve = vmsc
mem = int(math.floor(host_mem * overcommit / vmsc))
if os.popen("uname -i").readline().startswith("i386"):
logging.debug("Host is i386 architecture, max guest mem is 2G")
# Guest system with qemu overhead (64M) can't have more than 2G
if mem > 3100 - 64:
vmsc = int(math.ceil((host_mem * overcommit) /
(3100 - 64.0)))
if _host_reserve:
host_reserve += (vmsc - _host_reserve) * 64
host_mem -= (vmsc - _host_reserve) * 64
_host_reserve = vmsc
mem = int(math.floor(host_mem * overcommit / vmsc))
# 0.055 represents OS + TMPFS additional reserve per guest ram MB
if _guest_reserve:
guest_reserve += math.ceil(mem * 0.055)
swap = int(utils_memory.read_from_meminfo("SwapTotal")) / 1024
logging.debug("Overcommit = %f", overcommit)
logging.debug("True overcommit = %f ", (float(vmsc * mem) /
float(host_mem)))
logging.debug("Host memory = %dM", host_mem)
logging.debug("Guest memory = %dM", mem)
logging.debug("Using swap = %s", ksm_swap)
logging.debug("Swap = %dM", swap)
logging.debug("max_vms = %d", max_vms)
logging.debug("Count of all used VMs = %d", vmsc)
logging.debug("Performance_ratio = %f", perf_ratio)
# Generate unique keys for random series
skeys = []
dkeys = []
for i in range(0, max(vmsc, max_alloc)):
key = random.randrange(0, 255)
while key in skeys:
key = random.randrange(0, 255)
skeys.append(key)
key = random.randrange(0, 999)
while key in dkeys:
key = random.randrange(0, 999)
dkeys.append(key)
logging.debug("skeys: %s", skeys)
logging.debug("dkeys: %s", dkeys)
lvms = []
lsessions = []
# As we don't know the number and memory amount of VMs in advance,
# we need to specify and create them here
vm_name = params["main_vm"]
params['mem'] = mem
params['vms'] = vm_name
# Associate pidfile name
params['pid_' + vm_name] = utils_misc.generate_tmp_file_name(vm_name,
'pid')
if not params.get('extra_params'):
params['extra_params'] = ' '
params['extra_params_' + vm_name] = params.get('extra_params')
params['extra_params_' + vm_name] += (" -pidfile %s" %
(params.get('pid_' + vm_name)))
params['extra_params'] = params.get('extra_params_' + vm_name)
# ksm_size: amount of memory used by allocator
ksm_size = mem - guest_reserve
logging.debug("Memory used by allocator on guests = %dM", ksm_size)
fill_base_timeout = ksm_size / 10
# Creating the first guest
env_process.preprocess_vm(test, params, env, vm_name)
lvms.append(env.get_vm(vm_name))
if not lvms[0]:
test.error("VM object not found in environment")
if not lvms[0].is_alive():
test.error("VM seems to be dead; Test requires a living VM")
logging.debug("Booting first guest %s", lvms[0].name)
lsessions.append(lvms[0].wait_for_login(timeout=360))
# Associate vm PID
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
test.fail("Could not get PID of %s" % (vm_name))
# Creating other guest systems
for i in range(1, vmsc):
vm_name = "vm" + str(i + 1)
params['pid_' + vm_name] = utils_misc.generate_tmp_file_name(vm_name,
'pid')
params['extra_params_' + vm_name] = params.get('extra_params')
params['extra_params_' + vm_name] += (" -pidfile %s" %
(params.get('pid_' + vm_name)))
params['extra_params'] = params.get('extra_params_' + vm_name)
# Last VM is later used to run more allocators simultaneously
lvms.append(lvms[0].clone(vm_name, params))
env.register_vm(vm_name, lvms[i])
params['vms'] += " " + vm_name
logging.debug("Booting guest %s", lvms[i].name)
lvms[i].create()
if not lvms[i].is_alive():
test.error("VM %s seems to be dead; Test requires a"
"living VM" % lvms[i].name)
lsessions.append(lvms[i].wait_for_login(timeout=360))
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
test.fail("Could not get PID of %s" % (vm_name))
# Let guests rest a little bit :-)
pause = vmsc * 2 * perf_ratio
logging.debug("Waiting %ds before proceed", pause)
time.sleep(vmsc * 2 * perf_ratio)
logging.debug(utils_test.get_memory_info(lvms))
# Copy ksm_overcommit_guest.py into guests
vksmd_src = os.path.join(data_dir.get_shared_dir(),
"scripts", "ksm_overcommit_guest.py")
dst_dir = "/tmp"
for vm in lvms:
vm.copy_files_to(vksmd_src, dst_dir)
logging.info("Phase 0: PASS")
if params['ksm_mode'] == "parallel":
logging.info("Starting KSM test parallel mode")
split_parallel()
logging.info("KSM test parallel mode: PASS")
elif params['ksm_mode'] == "serial":
logging.info("Starting KSM test serial mode")
initialize_guests()
separate_first_guest()
split_guest()
logging.info("KSM test serial mode: PASS")
if not lvms[i].is_alive():
raise error.TestError("VM %s seems to be dead; Test requires a"
"living VM" % lvms[i].name)
lsessions.append(lvms[i].wait_for_login(timeout=360))
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
raise error.TestFail("Could not get PID of %s" % (vm_name))
# Let guests rest a little bit :-)
st = vmsc * 2 * perf_ratio
logging.debug("Waiting %ds before proceed", st)
time.sleep(vmsc * 2 * perf_ratio)
logging.debug(utils_test.get_memory_info(lvms))
# Copy ksm_overcommit_guest.py into guests
shared_dir = os.path.dirname(data_dir.get_data_dir())
vksmd_src = os.path.join(shared_dir, "scripts", "ksm_overcommit_guest.py")
dst_dir = "/tmp"
for vm in lvms:
vm.copy_files_to(vksmd_src, dst_dir)
logging.info("Phase 0: PASS")
if params['ksm_mode'] == "parallel":
logging.info("Starting KSM test parallel mode")
split_parallel()
logging.info("KSM test parallel mode: PASS")
elif params['ksm_mode'] == "serial":
logging.info("Starting KSM test serial mode")
def run(test, params, env):
"""
Tests KSM (Kernel Shared Memory) capability by allocating and filling
KVM guests memory using various values. KVM sets the memory as
MADV_MERGEABLE so all VM's memory can be merged. The workers in
guest writes to tmpfs filesystem thus allocations are not limited
by process max memory, only by VM's memory. Two test modes are supported -
serial and parallel.
Serial mode - uses multiple VMs, allocates memory per guest and always
verifies the correct number of shared memory.
0) Prints out the setup and initialize guest(s)
1) Fills guest with the same number (S1)
2) Random fill on the first guest
3) Random fill of the remaining VMs one by one until the
memory is completely filled (KVM stops machines which
asks for additional memory until there is available
memory) (S2, shouldn't finish)
4) Destroy all VMs but the last one
5) Checks the last VMs memory for corruption
Parallel mode - uses one VM with multiple allocator workers. Executes
scenarios in parallel to put more stress on the KVM.
0) Prints out the setup and initialize guest(s)
1) Fills memory with the same number (S1)
2) Fills memory with random numbers (S2)
3) Verifies all pages
4) Fills memory with the same number (S2)
5) Changes the last 96B (S3)
Scenarios:
S1) Fill all vms with the same value (all pages should be merged into 1)
S2) Random fill (all pages should be splitted)
S3) Fill last 96B (change only last 96B of each page; some pages will be
merged; there was a bug with data corruption)
Every worker has unique random key so we are able to verify the filled
values.
:param test: kvm test object.
:param params: Dictionary with test parameters.
:param env: Dictionary with the test environment.
:param cfg: ksm_swap - use swap?
:param cfg: ksm_overcommit_ratio - memory overcommit (serial mode only)
:param cfg: ksm_parallel_ratio - number of workers (parallel mode only)
:param cfg: ksm_host_reserve - override memory reserve on host in MB
:param cfg: ksm_guest_reserve - override memory reserve on guests in MB
:param cfg: ksm_mode - test mode {serial, parallel}
:param cfg: ksm_perf_ratio - performance ratio, increase it when your
machine is too slow
"""
def _start_allocator(vm, session, timeout):
"""
Execute ksm_overcommit_guest.py on guest, wait until it's initialized.
:param vm: VM object.
:param session: Remote session to a VM object.
:param timeout: Timeout that will be used to verify if
ksm_overcommit_guest.py started properly.
"""
logging.debug("Starting ksm_overcommit_guest.py on guest %s", vm.name)
session.sendline("python /tmp/ksm_overcommit_guest.py")
try:
session.read_until_last_line_matches(["PASS:"******"FAIL:"], timeout)
except aexpect.ExpectProcessTerminatedError as details:
e_msg = ("Command ksm_overcommit_guest.py on vm '%s' failed: %s" %
(vm.name, str(details)))
test.fail(e_msg)
def _execute_allocator(command, vm, session, timeout):
"""
Execute a given command on ksm_overcommit_guest.py main loop,
indicating the vm the command was executed on.
:param command: Command that will be executed.
:param vm: VM object.
:param session: Remote session to VM object.
:param timeout: Timeout used to verify expected output.
:return: Tuple (match index, data)
"""
logging.debug(
"Executing '%s' on ksm_overcommit_guest.py loop, "
"vm: %s, timeout: %s", command, vm.name, timeout)
session.sendline(command)
try:
(match,
data) = session.read_until_last_line_matches(["PASS:"******"FAIL:"],
timeout)
except aexpect.ExpectProcessTerminatedError as details:
e_msg = ("Failed to execute command '%s' on "
"ksm_overcommit_guest.py, vm '%s': %s" %
(command, vm.name, str(details)))
test.fail(e_msg)
return (match, data)
def get_ksmstat():
"""
Return sharing memory by ksm in MB
:return: memory in MB
"""
fpages = open('/sys/kernel/mm/ksm/pages_sharing')
ksm_pages = int(fpages.read())
fpages.close()
return ((ksm_pages * 4096) / 1e6)
def initialize_guests():
"""
Initialize guests (fill their memories with specified patterns).
"""
logging.info("Phase 1: filling guest memory pages")
for session in lsessions:
vm = lvms[lsessions.index(session)]
logging.debug("Turning off swap on vm %s", vm.name)
session.cmd("swapoff -a", timeout=300)
# Start the allocator
_start_allocator(vm, session, 60 * perf_ratio)
# Execute allocator on guests
for i in range(0, vmsc):
vm = lvms[i]
cmd = "mem = MemFill(%d, %s, %s)" % (ksm_size, skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % skeys[0]
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)
# Let ksm_overcommit_guest.py do its job
# (until shared mem reaches expected value)
shm = 0
j = 0
logging.debug("Target shared meminfo for guest %s: %s", vm.name,
ksm_size)
while ((new_ksm and (shm < (ksm_size * (i + 1))))
or (not new_ksm and (shm < (ksm_size)))):
if j > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until "
"the DL on guest: %s" % vm.name)
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug(
"Shared meminfo for guest %s after "
"iteration %s: %s", vm.name, j, shm)
j += 1
# Keep some reserve
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceeding...", pause)
time.sleep(pause)
logging.debug(utils_test.get_memory_info(lvms))
logging.info("Phase 1: PASS")
def separate_first_guest():
"""
Separate memory of the first guest by generating special random series
"""
logging.info("Phase 2: Split the pages on the first guest")
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, lvms[0], lsessions[0],
fill_base_timeout * 2 * perf_ratio)[1]
r_msg = data.splitlines()[-1]
logging.debug("Return message of static_random_fill: %s", r_msg)
out = int(r_msg.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s", ksm_size,
out, (ksm_size * 1000 / out))
logging.debug(utils_test.get_memory_info(lvms))
logging.debug("Phase 2: PASS")
def split_guest():
"""
Sequential split of pages on guests up to memory limit
"""
logging.info("Phase 3a: Sequential split of pages on guests up to "
"memory limit")
last_vm = 0
session = None
vm = None
for i in range(1, vmsc):
# Check VMs
for j in range(0, vmsc):
if not lvms[j].is_alive:
e_msg = ("VM %d died while executing static_random_fill on"
" VM %d in allocator loop" % (j, i))
test.fail(e_msg)
vm = lvms[i]
session = lsessions[i]
cmd = "mem.static_random_fill()"
logging.debug(
"Executing %s on ksm_overcommit_guest.py loop, "
"vm: %s", cmd, vm.name)
session.sendline(cmd)
out = ""
try:
logging.debug("Watching host mem while filling vm %s memory",
vm.name)
while (not out.startswith("PASS")
and not out.startswith("FAIL")):
if not vm.is_alive():
e_msg = ("VM %d died while executing "
"static_random_fill on allocator loop" % i)
test.fail(e_msg)
free_mem = int(utils_memory.read_from_meminfo("MemFree"))
if (ksm_swap):
free_mem = (
free_mem +
int(utils_memory.read_from_meminfo("SwapFree")))
logging.debug("Free memory on host: %d", free_mem)
# We need to keep some memory for python to run.
if (free_mem < 64000) or (ksm_swap and free_mem <
(450000 * perf_ratio)):
vm.pause()
for j in range(0, i):
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
logging.debug("Only %s free memory, killing %d guests",
free_mem, (i - 1))
last_vm = i
out = session.read_nonblocking(0.1, 1)
time.sleep(2)
except OSError:
logging.debug("Only %s host free memory, killing %d guests",
free_mem, (i - 1))
logging.debug("Stopping %s", vm.name)
vm.pause()
for j in range(0, i):
logging.debug("Destroying %s", lvms[j].name)
lvms[j].destroy(gracefully=False)
time.sleep(20)
vm.resume()
last_vm = i
if last_vm != 0:
break
logging.debug("Memory filled for guest %s", vm.name)
logging.info("Phase 3a: PASS")
logging.info("Phase 3b: Verify memory of the max stressed VM")
for i in range(last_vm + 1, vmsc):
lsessions[i].close()
if i == (vmsc - 1):
logging.debug(utils_test.get_memory_info([lvms[i]]))
logging.debug("Destroying guest %s", lvms[i].name)
lvms[i].destroy(gracefully=False)
# Verify last machine with randomly generated memory
cmd = "mem.static_random_verify()"
_execute_allocator(cmd, lvms[last_vm], lsessions[last_vm],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([lvms[last_vm]]))
lsessions[last_vm].cmd_output("die()", 20)
lvms[last_vm].destroy(gracefully=False)
logging.info("Phase 3b: PASS")
def split_parallel():
"""
Parallel page spliting
"""
logging.info("Phase 1: parallel page spliting")
# We have to wait until allocator is finished (it waits 5 seconds to
# clean the socket
session = lsessions[0]
vm = lvms[0]
for i in range(1, max_alloc):
lsessions.append(vm.wait_for_login(timeout=360))
session.cmd("swapoff -a", timeout=300)
for i in range(0, max_alloc):
# Start the allocator
_start_allocator(vm, lsessions[i], 60 * perf_ratio)
logging.info("Phase 1: PASS")
logging.info("Phase 2a: Simultaneous merging")
logging.debug("Memory used by allocator on guests = %dMB",
(ksm_size / max_alloc))
for i in range(0, max_alloc):
cmd = "mem = MemFill(%d, %s, %s)" % (
(ksm_size / max_alloc), skeys[i], dkeys[i])
_execute_allocator(cmd, vm, lsessions[i], 60 * perf_ratio)
cmd = "mem.value_fill(%d)" % (skeys[0])
_execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)
# Wait until ksm_overcommit_guest.py merges pages (3 * ksm_size / 3)
shm = 0
i = 0
logging.debug("Target shared memory size: %s", ksm_size)
while (shm < ksm_size):
if i > 64:
logging.debug(utils_test.get_memory_info(lvms))
test.error("SHM didn't merge the memory until DL")
pause = ksm_size / 200 * perf_ratio
logging.debug("Waiting %ds before proceed...", pause)
time.sleep(pause)
if (new_ksm):
shm = get_ksmstat()
else:
shm = vm.get_shared_meminfo()
logging.debug("Shared meminfo after attempt %s: %s", i, shm)
i += 1
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2a: PASS")
logging.info("Phase 2b: Simultaneous spliting")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.static_random_fill()"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
logging.debug(data)
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
(ksm_size / max_alloc), out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2b: PASS")
logging.info("Phase 2c: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify()"
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2c: PASS")
logging.info("Phase 2d: Simultaneous merging")
# Actual splitting
for i in range(0, max_alloc):
cmd = "mem.value_fill(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * 2 * perf_ratio)[1]
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2d: PASS")
logging.info("Phase 2e: Simultaneous verification")
for i in range(0, max_alloc):
cmd = "mem.value_check(%d)" % skeys[0]
data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))[1]
logging.info("Phase 2e: PASS")
logging.info("Phase 2f: Simultaneous spliting last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_fill(96)"
data = _execute_allocator(cmd, vm, lsessions[i],
fill_base_timeout * perf_ratio)[1]
data = data.splitlines()[-1]
out = int(data.split()[4])
logging.debug("Performance: %dMB * 1000 / %dms = %dMB/s",
ksm_size / max_alloc, out,
(ksm_size * 1000 / out / max_alloc))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2f: PASS")
logging.info("Phase 2g: Simultaneous verification last 96B")
for i in range(0, max_alloc):
cmd = "mem.static_random_verify(96)"
_, data = _execute_allocator(cmd, vm, lsessions[i],
(mem / 200 * 50 * perf_ratio))
logging.debug(utils_test.get_memory_info([vm]))
logging.info("Phase 2g: PASS")
logging.debug("Cleaning up...")
for i in range(0, max_alloc):
lsessions[i].cmd_output("die()", 20)
session.close()
vm.destroy(gracefully=False)
# Main test code
logging.info("Starting phase 0: Initialization")
if process.run("ps -C ksmtuned", ignore_status=True).exit_status == 0:
logging.info("Killing ksmtuned...")
process.run("killall ksmtuned")
new_ksm = False
if (os.path.exists("/sys/kernel/mm/ksm/run")):
process.run("echo 50 > /sys/kernel/mm/ksm/sleep_millisecs", shell=True)
process.run("echo 5000 > /sys/kernel/mm/ksm/pages_to_scan", shell=True)
process.run("echo 1 > /sys/kernel/mm/ksm/run", shell=True)
e_up = "/sys/kernel/mm/transparent_hugepage/enabled"
e_rh = "/sys/kernel/mm/redhat_transparent_hugepage/enabled"
if os.path.exists(e_up):
process.run("echo 'never' > %s" % e_up, shell=True)
if os.path.exists(e_rh):
process.run("echo 'never' > %s" % e_rh, shell=True)
new_ksm = True
else:
try:
process.run("modprobe ksm")
process.run("ksmctl start 5000 100")
except process.CmdError as details:
test.fail("Failed to load KSM: %s" % details)
# host_reserve: mem reserve kept for the host system to run
host_reserve = int(params.get("ksm_host_reserve", -1))
if (host_reserve == -1):
try:
available = utils_memory.read_from_meminfo("MemAvailable")
except process.CmdError: # ancient kernels
utils_memory.drop_caches()
available = utils_memory.read_from_meminfo("MemFree")
# default host_reserve = UsedMem + one_minimal_guest(128MB)
# later we add 64MB per additional guest
host_reserve = ((utils_memory.memtotal() - available) / 1024 + 128)
# using default reserve
_host_reserve = True
else:
_host_reserve = False
# guest_reserve: mem reserve kept to avoid guest OS to kill processes
guest_reserve = int(params.get("ksm_guest_reserve", -1))
if (guest_reserve == -1):
# default guest_reserve = minimal_system_mem(256MB)
# later we add tmpfs overhead
guest_reserve = 256
# using default reserve
_guest_reserve = True
else:
_guest_reserve = False
max_vms = int(params.get("max_vms", 2))
overcommit = float(params.get("ksm_overcommit_ratio", 2.0))
max_alloc = int(params.get("ksm_parallel_ratio", 1))
# vmsc: count of all used VMs
vmsc = int(overcommit) + 1
vmsc = max(vmsc, max_vms)
if (params['ksm_mode'] == "serial"):
max_alloc = vmsc
if _host_reserve:
# First round of additional guest reserves
host_reserve += vmsc * 64
_host_reserve = vmsc
host_mem = (int(utils_memory.memtotal()) / 1024 - host_reserve)
ksm_swap = False
if params.get("ksm_swap") == "yes":
ksm_swap = True
# Performance ratio
perf_ratio = params.get("ksm_perf_ratio")
if perf_ratio:
perf_ratio = float(perf_ratio)
else:
perf_ratio = 1
if (params['ksm_mode'] == "parallel"):
vmsc = 1
overcommit = 1
mem = host_mem
# 32bit system adjustment
if "64" not in params.get("vm_arch_name"):
logging.debug("Probably i386 guest architecture, "
"max allocator mem = 2G")
# Guest can have more than 2G but
# kvm mem + 1MB (allocator itself) can't
if (host_mem > 3100):
mem = 3100
if os.popen("uname -i").readline().startswith("i386"):
logging.debug("Host is i386 architecture, max guest mem is 2G")
# Guest system with qemu overhead (64M) can't have more than 2G
if mem > 3100 - 64:
mem = 3100 - 64
else:
# mem: Memory of the guest systems. Maximum must be less than
# host's physical ram
mem = int(overcommit * host_mem / vmsc)
# 32bit system adjustment
if not params['image_name'].endswith("64"):
logging.debug("Probably i386 guest architecture, "
"max allocator mem = 2G")
# Guest can have more than 2G but
# kvm mem + 1MB (allocator itself) can't
if mem - guest_reserve - 1 > 3100:
vmsc = int(
math.ceil(
(host_mem * overcommit) / (3100 + guest_reserve)))
if _host_reserve:
host_reserve += (vmsc - _host_reserve) * 64
host_mem -= (vmsc - _host_reserve) * 64
_host_reserve = vmsc
mem = int(math.floor(host_mem * overcommit / vmsc))
if os.popen("uname -i").readline().startswith("i386"):
logging.debug("Host is i386 architecture, max guest mem is 2G")
# Guest system with qemu overhead (64M) can't have more than 2G
if mem > 3100 - 64:
vmsc = int(math.ceil((host_mem * overcommit) / (3100 - 64.0)))
if _host_reserve:
host_reserve += (vmsc - _host_reserve) * 64
host_mem -= (vmsc - _host_reserve) * 64
_host_reserve = vmsc
mem = int(math.floor(host_mem * overcommit / vmsc))
# 0.055 represents OS + TMPFS additional reserve per guest ram MB
if _guest_reserve:
guest_reserve += math.ceil(mem * 0.055)
swap = int(utils_memory.read_from_meminfo("SwapTotal")) / 1024
logging.debug("Overcommit = %f", overcommit)
logging.debug("True overcommit = %f ",
(float(vmsc * mem) / float(host_mem)))
logging.debug("Host memory = %dM", host_mem)
logging.debug("Guest memory = %dM", mem)
logging.debug("Using swap = %s", ksm_swap)
logging.debug("Swap = %dM", swap)
logging.debug("max_vms = %d", max_vms)
logging.debug("Count of all used VMs = %d", vmsc)
logging.debug("Performance_ratio = %f", perf_ratio)
# Generate unique keys for random series
skeys = []
dkeys = []
for i in range(0, max(vmsc, max_alloc)):
key = random.randrange(0, 255)
while key in skeys:
key = random.randrange(0, 255)
skeys.append(key)
key = random.randrange(0, 999)
while key in dkeys:
key = random.randrange(0, 999)
dkeys.append(key)
logging.debug("skeys: %s", skeys)
logging.debug("dkeys: %s", dkeys)
lvms = []
lsessions = []
# As we don't know the number and memory amount of VMs in advance,
# we need to specify and create them here
vm_name = params["main_vm"]
params['mem'] = mem
params['vms'] = vm_name
# Associate pidfile name
params['pid_' + vm_name] = utils_misc.generate_tmp_file_name(
vm_name, 'pid')
if not params.get('extra_params'):
params['extra_params'] = ' '
params['extra_params_' + vm_name] = params.get('extra_params')
params['extra_params_' + vm_name] += (" -pidfile %s" %
(params.get('pid_' + vm_name)))
params['extra_params'] = params.get('extra_params_' + vm_name)
# ksm_size: amount of memory used by allocator
ksm_size = mem - guest_reserve
logging.debug("Memory used by allocator on guests = %dM", ksm_size)
fill_base_timeout = ksm_size / 10
# Creating the first guest
env_process.preprocess_vm(test, params, env, vm_name)
lvms.append(env.get_vm(vm_name))
if not lvms[0]:
test.error("VM object not found in environment")
if not lvms[0].is_alive():
test.error("VM seems to be dead; Test requires a living VM")
logging.debug("Booting first guest %s", lvms[0].name)
lsessions.append(lvms[0].wait_for_login(timeout=360))
# Associate vm PID
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
test.fail("Could not get PID of %s" % (vm_name))
# Creating other guest systems
for i in range(1, vmsc):
vm_name = "vm" + str(i + 1)
params['pid_' + vm_name] = utils_misc.generate_tmp_file_name(
vm_name, 'pid')
params['extra_params_' + vm_name] = params.get('extra_params')
params['extra_params_' + vm_name] += (" -pidfile %s" %
(params.get('pid_' + vm_name)))
params['extra_params'] = params.get('extra_params_' + vm_name)
# Last VM is later used to run more allocators simultaneously
lvms.append(lvms[0].clone(vm_name, params))
env.register_vm(vm_name, lvms[i])
params['vms'] += " " + vm_name
logging.debug("Booting guest %s", lvms[i].name)
lvms[i].create()
if not lvms[i].is_alive():
test.error("VM %s seems to be dead; Test requires a"
"living VM" % lvms[i].name)
lsessions.append(lvms[i].wait_for_login(timeout=360))
try:
tmp = open(params.get('pid_' + vm_name), 'r')
params['pid_' + vm_name] = int(tmp.readline())
except Exception:
test.fail("Could not get PID of %s" % (vm_name))
# Let guests rest a little bit :-)
pause = vmsc * 2 * perf_ratio
logging.debug("Waiting %ds before proceed", pause)
time.sleep(vmsc * 2 * perf_ratio)
logging.debug(utils_test.get_memory_info(lvms))
# Copy ksm_overcommit_guest.py into guests
vksmd_src = os.path.join(data_dir.get_shared_dir(), "scripts",
"ksm_overcommit_guest.py")
dst_dir = "/tmp"
for vm in lvms:
vm.copy_files_to(vksmd_src, dst_dir)
logging.info("Phase 0: PASS")
if params['ksm_mode'] == "parallel":
logging.info("Starting KSM test parallel mode")
split_parallel()
logging.info("KSM test parallel mode: PASS")
elif params['ksm_mode'] == "serial":
logging.info("Starting KSM test serial mode")
initialize_guests()
separate_first_guest()
split_guest()
logging.info("KSM test serial mode: PASS")
