Re: [PATCH v2] Netperf: add option udp_size

On Tue, Jul 19, 2016 at 04:49:57PM +0200, Kamil Jerabek wrote:

This option set UDP datagram size for netperf test on client. If this option is not explicitly specified, default value from netperf is used.

The option accepts integer value describing datagram size, it should be also concatenated with G/M/K/g/m/k.

In this commit I also added this option to regression_test: simple_netperf test.

recipes/regression_tests/phase1/simple_netperf.py | 10 ++++++++-- test_modules/Netperf.py | 11 +++++++++++ 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/recipes/regression_tests/phase1/simple_netperf.py b/recipes/regression_tests/phase1/simple_netperf.py index fe9d96b..d68d10c 100644 --- a/recipes/regression_tests/phase1/simple_netperf.py +++ b/recipes/regression_tests/phase1/simple_netperf.py @@ -43,6 +43,7 @@ nperf_mode = ctl.get_alias("nperf_mode") nperf_num_parallel = int(ctl.get_alias("nperf_num_parallel")) nperf_debug = ctl.get_alias("nperf_debug") nperf_max_dev = ctl.get_alias("nperf_max_dev") +nperf_udp_size = ctl.get_alias("nperf_udp_size") pr_user_comment = ctl.get_alias("perfrepo_comment")

pr_comment = generate_perfrepo_comment([m1, m2], pr_user_comment) @@ -107,7 +108,8 @@ netperf_cli_udp = ctl.get_module("Netperf", "runs" : nperf_max_runs, "netperf_opts" : p_opts, "debug" : nperf_debug,

•                                  "max_deviation" : nperf_max_dev
  

•                                  "max_deviation" : nperf_max_dev,
  

•                                  "udp_size" : nperf_udp_size
                               })
  
  

netperf_cli_udp6 = ctl.get_module("Netperf", @@ -121,7 +123,8 @@ netperf_cli_udp6 = ctl.get_module("Netperf", "runs" : nperf_max_runs, "netperf_opts" : p_opts6, "debug" : nperf_debug,

•                                  "max_deviation" : nperf_max_dev
  

•                                  "max_deviation" : nperf_max_dev,
  

•                                  "udp_size" : nperf_udp_size
                               })
  
  

netperf_srv = ctl.get_module("Netperf", @@ -198,6 +201,9 @@ for setting in offload_settings: for offload in setting: result_udp.set_parameter(offload[0], offload[1])

•    if nperf_udp_size is not None:
  

•        result_udp.set_parameter("nperf_udp_size", nperf_udp_size)
  

•    result_udp.add_tag(product_name)
     if nperf_mode == "multi":
         result_udp.add_tag("multithreaded")
  

diff --git a/test_modules/Netperf.py b/test_modules/Netperf.py index d09bdfc..e6c68ad 100644 --- a/test_modules/Netperf.py +++ b/test_modules/Netperf.py @@ -39,6 +39,7 @@ class Netperf(TestGeneric): self._cpu_util = self.get_opt("cpu_util") self._num_parallel = int(self.get_opt("num_parallel", default=1)) self._runs = self.get_opt("runs", default=1)

•    self._udp_size = self.get_opt("udp_size")
     self._debug = int_it(self.get_opt("debug", default=0))
  
     self._threshold = self._parse_threshold(self.get_opt("threshold"))
  

@@ -141,6 +142,16 @@ class Netperf(TestGeneric): else: cmd += " -- %s" % self._testoptions

•        if self._udp_size is not None:
  

•            """
  

•            udp packets will have this size
  

•            """
  

•            if self._is_omni() or self._testoptions:
  

•                cmd += " -m %s" % self._udp_size
  

•            else:
  

•                cms += " -- -m %s" % self._udp_size
  

•    elif self._role == "server":
         cmd = "netserver -D"
         if self._bind is not None:
  

-- 2.5.5 _______________________________________________ LNST-developers mailing list lnst-developers@lists.fedorahosted.org https://lists.fedorahosted.org/admin/lists/lnst-developers@lists.fedorahoste...

Unfortunatelly, LNST currently has this bug: https://github.com/jpirko/lnst/issues/158

So when the "udp_size" alias isn't specified, None is used and the Netperf module receives it as a string 'None' instead of a normal None object... So we end up with a netperf command that contains '-m None'.

Fortunatelly this doesn't cause Netperf to crash, and it will solve itself on its own once I fix the linked issue, but that will have to wait for Pyrecipes. Until then I still think this should be avoided.

I guess the most reasonable suggestion I can give is to set the option using the "update_options()" method call in an if statement, kind of like what we do with nperf_mode.

A couple of notes that I've already told you in person but just so we have it on record: * Wrap the commit description to a reasonable line length - 80 chars * don't forget the Signed-off-by line...

-Ondrej