diff options
author | Peter Mikus <pmikus@cisco.com> | 2019-12-19 19:20:20 +0000 |
---|---|---|
committer | Peter Mikus <pmikus@cisco.com> | 2020-01-20 09:01:31 +0000 |
commit | 2b3a18b05be40ceda745c74ec655ef548100d769 (patch) | |
tree | 485e03fa1f555687821555456b0ee55547578765 /resources/libraries/python | |
parent | 87f81f123c054908df6770f11a947b9fe6ab0e93 (diff) |
FIX: Hugepages should be detected also from overcommit
Change-Id: I36ec9d61f775791ad0632a579c4921889d1f5acf
Signed-off-by: Peter Mikus <pmikus@cisco.com>
Diffstat (limited to 'resources/libraries/python')
-rw-r--r-- | resources/libraries/python/DUTSetup.py | 216 | ||||
-rw-r--r-- | resources/libraries/python/QemuUtils.py | 2 |
2 files changed, 70 insertions, 148 deletions
diff --git a/resources/libraries/python/DUTSetup.py b/resources/libraries/python/DUTSetup.py index 431ccfb8ae..3f8c407313 100644 --- a/resources/libraries/python/DUTSetup.py +++ b/resources/libraries/python/DUTSetup.py @@ -754,176 +754,98 @@ class DUTSetup: return stdout.strip() @staticmethod - def get_huge_page_size(node): - """Get default size of huge pages in system. + def get_hugepages_info(node, hugesize=None): + """Get number of huge pages in system. :param node: Node in the topology. + :param hugesize: Size of hugepages. Default system huge size if None. :type node: dict - :returns: Default size of free huge pages in system. - :rtype: int - :raises RuntimeError: If reading failed for three times. - """ - ssh = SSH() - ssh.connect(node) - - for _ in range(3): - ret_code, stdout, _ = ssh.exec_command_sudo( - u"grep Hugepagesize /proc/meminfo | awk '{ print $2 }'" - ) - if ret_code == 0: - try: - huge_size = int(stdout) - except ValueError: - logger.trace(u"Reading huge page size information failed") - else: - break - else: - raise RuntimeError(u"Getting huge page size information failed.") - return huge_size - - @staticmethod - def get_huge_page_free(node, huge_size): - """Get number of free huge pages in system. - - :param node: Node in the topology. - :param huge_size: Size of hugepages. - :type node: dict - :type huge_size: int - :returns: Number of free huge pages in system. - :rtype: int - :raises RuntimeError: If reading failed for three times. - """ - # TODO: add numa aware option - ssh = SSH() - ssh.connect(node) - - for _ in range(3): - ret_code, stdout, _ = ssh.exec_command_sudo( - f"cat /sys/kernel/mm/hugepages/hugepages-{huge_size}kB/" - f"free_hugepages" - ) - if ret_code == 0: - try: - huge_free = int(stdout) - except ValueError: - logger.trace(u"Reading free huge pages information failed") - else: - break - else: - raise RuntimeError(u"Getting free huge pages information failed.") - return huge_free - - @staticmethod - def get_huge_page_total(node, huge_size): - """Get total number of huge pages in system. - - :param node: Node in the topology. - :param huge_size: Size of hugepages. - :type node: dict - :type huge_size: int - :returns: Total number of huge pages in system. - :rtype: int - :raises RuntimeError: If reading failed for three times. + :type hugesize: int + :returns: Number of huge pages in system. + :rtype: dict + :raises RuntimeError: If reading failed. """ - # TODO: add numa aware option - ssh = SSH() - ssh.connect(node) - - for _ in range(3): - ret_code, stdout, _ = ssh.exec_command_sudo( - f"cat /sys/kernel/mm/hugepages/hugepages-{huge_size}kB/" - f"nr_hugepages" - ) - if ret_code == 0: - try: - huge_total = int(stdout) - except ValueError: - logger.trace(u"Reading total huge pages information failed") - else: - break - else: - raise RuntimeError(u"Getting total huge pages information failed.") - return huge_total - - @staticmethod - def check_huge_page(node, huge_mnt, mem_size, allocate=False): + if not hugesize: + hugesize = "$(grep Hugepagesize /proc/meminfo | awk '{ print $2 }')" + command = f"cat /sys/kernel/mm/hugepages/hugepages-{hugesize}kB/*" + stdout, _ = exec_cmd_no_error(node, command) + try: + line = stdout.splitlines() + return { + "free_hugepages": int(line[0]), + "nr_hugepages": int(line[1]), + "nr_hugepages_mempolicy": int(line[2]), + "nr_overcommit_hugepages": int(line[3]), + "resv_hugepages": int(line[4]), + "surplus_hugepages": int(line[5]) + } + except ValueError: + logger.trace(u"Reading huge pages information failed!") + + @staticmethod + def check_huge_page( + node, huge_mnt, mem_size, hugesize=2048, allocate=False): """Check if there is enough HugePages in system. If allocate is set to true, try to allocate more HugePages. :param node: Node in the topology. :param huge_mnt: HugePage mount point. - :param mem_size: Requested memory in MB. + :param mem_size: Reqeusted memory in MB. + :param hugesize: HugePage size in KB. :param allocate: Whether to allocate more memory if not enough. :type node: dict :type huge_mnt: str - :type mem_size: str + :type mem_size: int + :type hugesize: int :type allocate: bool :raises RuntimeError: Mounting hugetlbfs failed or not enough HugePages or increasing map count failed. """ - # TODO: split function into smaller parts. - ssh = SSH() - ssh.connect(node) + # Get huge pages information. + hugepages = DUTSetup.get_hugepages_info(node, hugesize=hugesize) - # Get huge pages information - huge_size = DUTSetup.get_huge_page_size(node) - huge_free = DUTSetup.get_huge_page_free(node, huge_size) - huge_total = DUTSetup.get_huge_page_total(node, huge_size) + # Check if hugepages requested are available on node. + if hugepages[u"nr_overcommit_hugepages"]: + # If overcommit is used, we need to know how many additional pages + # we can allocate + huge_available = hugepages[u"nr_overcommit_hugepages"] - \ + hugepages[u"surplus_hugepages"] + else: + # Fallbacking to free_hugepages which were used before to detect. + huge_available = hugepages[u"free_hugepages"] - # Check if memory requested is available on - mem_size = int(mem_size) - if (mem_size * 1024) > (huge_free * huge_size): - # If we want to allocate hugepage dynamically + if ((mem_size * 1024) // hugesize) > huge_available: + # If we want to allocate hugepage dynamically. if allocate: - mem_needed = (mem_size * 1024) - (huge_free * huge_size) - huge_to_allocate = ((mem_needed // huge_size) * 2) + huge_total - max_map_count = huge_to_allocate*4 - # Increase maximum number of memory map areas a process may have - ret_code, _, _ = ssh.exec_command_sudo( + huge_needed = ((mem_size * 1024) // hugesize) - huge_available + huge_to_allocate = huge_needed + hugepages[u"nr_hugepages"] + max_map_count = huge_to_allocate * 4 + # Check if huge pages mount point exist. + try: + exec_cmd_no_error(node, u"fgrep 'hugetlbfs' /proc/mounts") + except RuntimeError: + exec_cmd_no_error(node, f"mkdir -p {huge_mnt}", sudo=True) + exec_cmd_no_error( + node, + f"mount -t hugetlbfs -o pagesize={hugesize}k none " + f"{huge_mnt}", + sudo=True) + # Increase maximum number of memory map areas for process. + exec_cmd_no_error( + node, f"echo \"{max_map_count}\" | " - f"sudo tee /proc/sys/vm/max_map_count" + f"sudo tee /proc/sys/vm/max_map_count", + message=f"Increase map count failed on {node[u'host']}!" ) - if int(ret_code) != 0: - raise RuntimeError( - f"Increase map count failed on {node[u'host']}" - ) - # Increase hugepage count - ret_code, _, _ = ssh.exec_command_sudo( + # Increase hugepage count. + exec_cmd_no_error( + node, f"echo \"{huge_to_allocate}\" | " - f"sudo tee /proc/sys/vm/nr_hugepages" + f"sudo tee /proc/sys/vm/nr_hugepages", + message=f"Mount huge pages failed on {node[u'host']}!" ) - if int(ret_code) != 0: - raise RuntimeError( - f"Mount huge pages failed on {node[u'host']}" - ) - # If we do not want to allocate dynamically end with error + # If we do not want to allocate dynamically end with error. else: raise RuntimeError( - f"Not enough free huge pages: {huge_free}, " - f"{huge_free * huge_size} MB" - ) - # Check if huge pages mount point exist - has_huge_mnt = False - ret_code, stdout, _ = ssh.exec_command(u"cat /proc/mounts") - if int(ret_code) == 0: - for line in stdout.splitlines(): - # Try to find something like: - # none /mnt/huge hugetlbfs rw,realtime,pagesize=2048k 0 0 - mount = line.split() - if mount[2] == u"hugetlbfs" and mount[1] == huge_mnt: - has_huge_mnt = True - break - # If huge page mount point not exist create one - if not has_huge_mnt: - ret_code, _, _ = ssh.exec_command_sudo(f"mkdir -p {huge_mnt}") - if int(ret_code) != 0: - raise RuntimeError( - f"Create mount dir failed on {node[u'host']}" - ) - ret_code, _, _ = ssh.exec_command_sudo( - f"mount -t hugetlbfs -o pagesize=2048k none {huge_mnt}" - ) - if int(ret_code) != 0: - raise RuntimeError( - f"Mount huge pages failed on {node[u'host']}" + f"Not enough availablehuge pages: {huge_available}!" ) diff --git a/resources/libraries/python/QemuUtils.py b/resources/libraries/python/QemuUtils.py index f31e153bd2..2b08429d84 100644 --- a/resources/libraries/python/QemuUtils.py +++ b/resources/libraries/python/QemuUtils.py @@ -611,7 +611,7 @@ class QemuUtils: message = f"QEMU: Start failed on {self._node[u'host']}!" try: DUTSetup.check_huge_page( - self._node, u"/dev/hugepages", self._opt.get(u"mem")) + self._node, u"/dev/hugepages", int(self._opt.get(u"mem"))) exec_cmd_no_error( self._node, cmd_opts, timeout=300, sudo=True, message=message |