feat(telemetry): Add telemetry export

Signed-off-by: pmikus <peter.mikus@protonmail.ch>
Change-Id: Ibb8a59fd4e08bc736494c7ab18e689a55f7a98dc

4 files changed, 66 insertions, 92 deletions

diff --git a/resources/libraries/python/TelemetryUtil.py b/resources/libraries/python/TelemetryUtil.py
index c978441d72..70efa13fff 100644
--- a/resources/libraries/python/TelemetryUtil.py
+++ b/resources/libraries/python/TelemetryUtil.py
@@ -13,131 +13,87 @@
 
 """Telemetry utility."""
 
-from robot.api import logger
-from time import sleep
-
+from resources.libraries.python.model.ExportResult import append_telemetry
 from resources.libraries.python.Constants import Constants
-from resources.libraries.python.VppCounters import VppCounters
-from resources.libraries.python.OptionString import OptionString
-from resources.libraries.python.ssh import exec_cmd, exec_cmd_no_error
+from resources.libraries.python.ssh import exec_cmd_no_error
 from resources.libraries.python.topology import NodeType
 
-__all__ = [u"TelemetryUtil"]
+__all__ = ["TelemetryUtil"]
 
 
 class TelemetryUtil:
     """Class contains methods for telemetry utility."""
 
     @staticmethod
-    def perf_stat(node, cpu_list=None, duration=1):
-        """Get perf stat read for duration.
-
-        :param node: Node in the topology.
-        :param cpu_list: CPU List as a string separated by comma.
-        :param duration: Measure time in seconds.
-        :type node: dict
-        :type cpu_list: str
-        :type duration: int
-        """
-        if cpu_list:
-            cpu_list = list(dict.fromkeys(cpu_list.split(u",")))
-            cpu_list = ",".join(str(cpu) for cpu in cpu_list)
-
-        cmd_opts = OptionString(prefix=u"--")
-        cmd_opts.add(u"no-aggr")
-        cmd_opts.add_with_value_if(
-            u"cpu", cpu_list, cpu_list
-        )
-        cmd_opts.add_if(
-            u"all-cpus", not(cpu_list)
-        )
-        cmd_opts.add_with_value_if(
-            u"event", f"'{{{Constants.PERF_STAT_EVENTS}}}'",
-            Constants.PERF_STAT_EVENTS
-        )
-        cmd_opts.add_with_value(
-            u"interval-print", 1000
-        )
-        cmd_opts.add_with_value(
-            u"field-separator", u"';'"
-        )
-
-        cmd_base = OptionString()
-        cmd_base.add(f"perf stat")
-        cmd_base.extend(cmd_opts)
-        cmd_base.add(u"--")
-        cmd_base.add_with_value(u"sleep", int(duration))
-
-        exec_cmd(node, cmd_base, sudo=True)
-
-    @staticmethod
-    def perf_stat_on_all_duts(nodes, cpu_list=None, duration=1):
-        """Get perf stat read for duration on all DUTs.
-
-        :param nodes: Nodes in the topology.
-        :param cpu_list: CPU List.
-        :param duration: Measure time in seconds.
-        :type nodes: dict
-        :type cpu_list: str
-        :type duration: int
-        """
-        for node in nodes.values():
-            if node[u"type"] == NodeType.DUT:
-                TelemetryUtil.perf_stat(
-                    node, cpu_list=cpu_list, duration=duration
-                )
-
-    @staticmethod
-    def run_telemetry(node, profile, hook=None):
-        """Get telemetry stat read for duration.
+    def _run_telemetry(
+            node, profile, sid=None, spath=None, oload="", export=False):
+        """Get telemetry read on node.
 
         :param node: Node in the topology.
         :param profile: Telemetry configuration profile.
-        :param hook: Process ID or socket path (optional).
+        :param sid: Socket ID used to describe recipient side of socket.
+        :param spath: Socket path.
+        :param oload: Telemetry offered load, unique within the test (optional).
+        :param export: If false, do not attempt JSON export (default false).
         :type node: dict
         :type profile: str
-        :type hook: str
+        :type sid: str
+        :type spath: str
+        :type oload: str
+        :type export: bool
         """
-        config = u""
+        config = ""
         config += f"{Constants.REMOTE_FW_DIR}/"
         config += f"{Constants.RESOURCES_TPL_TELEMETRY}/"
         config += f"{profile}"
 
-        cd_cmd = u""
+        cd_cmd = ""
         cd_cmd += f"sh -c \"cd {Constants.REMOTE_FW_DIR}/"
         cd_cmd += f"{Constants.RESOURCES_TOOLS}"
 
-        bin_cmd = f"python3 -m telemetry --config {config} --hook {hook}\""
-        hostname = node[u"host"]
-
+        if spath:
+            bin_cmd = f"python3 -m telemetry --config {config} --hook {spath}\""
+        else:
+            bin_cmd = f"python3 -m telemetry --config {config}\""
         exec_cmd_no_error(node, f"{cd_cmd} && {bin_cmd}", sudo=True)
+
+        if not export:
+            return
+
+        hostname = exec_cmd_no_error(node, "hostname")[0].strip()
         stdout, _ = exec_cmd_no_error(
-            node, u"cat /tmp/metric.prom", sudo=True, log_stdout_err=False
-        )
-        logger.info(
-            u"# TYPE target info\n"
-            u"# HELP target Target metadata\n"
-            f"target_info{{hostname=\"{hostname}\",hook=\"{hook}\"}} 1\n"
-            f"{stdout}"
+            node, "cat /tmp/metric.prom", sudo=True, log_stdout_err=False
         )
+        prefix = "{"
+        prefix += f"hostname=\"{hostname}\","
+        if sid:
+            prefix += f"hook=\"{sid}\","
+        prefix += f"oload=\"{oload}\","
+        for line in stdout.splitlines():
+            if line and not line.startswith("#"):
+                append_telemetry(
+                    prefix.join(line.rsplit("{", 1)).replace("\"", "'")
+                )
 
-    @staticmethod
-    def run_telemetry_on_all_duts(nodes, profile):
-        """Get telemetry stat read on all DUTs.
+    def run_telemetry_on_all_duts(self, nodes, profile, oload="", export=False):
+        """Get telemetry read on all DUTs.
 
         :param nodes: Nodes in the topology.
         :param profile: Telemetry configuration profile.
-        :param hooks: Dict of Process IDs or socket paths (optional).
+        :param oload: Telemetry offered load, unique within the test (optional).
+        :param export: If false, do not attempt JSON export (default false).
         :type nodes: dict
         :type profile: str
-        :type hooks: dict
+        :type oload: str
+        :type export: bool
         """
         for node in nodes.values():
-            if node[u"type"] == NodeType.DUT:
+            if node["type"] == NodeType.DUT:
                 try:
-                    for socket in node[u"sockets"][u"CLI"].values():
-                        TelemetryUtil.run_telemetry(
-                            node, profile=profile, hook=socket
+                    for sid, spath in node["sockets"]["CLI"].items():
+                        self._run_telemetry(
+                            node, profile=profile, sid=sid, spath=spath,
+                            oload=oload, export=export
                         )
                 except IndexError:
                     pass
diff --git a/resources/libraries/python/model/ExportResult.py b/resources/libraries/python/model/ExportResult.py
index dbe2914565..31840c9deb 100644
--- a/resources/libraries/python/model/ExportResult.py
+++ b/resources/libraries/python/model/ExportResult.py
@@ -210,3 +210,12 @@ def export_ndrpdr_latency(text, latency):
     if len(latency) < 2:
         return
     _add_latency(result_node, percent, u"reverse", latency[1])
+
+def append_telemetry(telemetry_item):
+    """Append telemetry entry to proper place so it is dumped into json.
+
+    :param telemetry_item: Telemetry entry.
+    :type telemetry_item: str
+    """
+    data = get_export_data()
+    data[u"telemetry"].append(telemetry_item)
diff --git a/resources/libraries/robot/performance/performance_actions.robot b/resources/libraries/robot/performance/performance_actions.robot
index 40f0bc9999..0404ce2a52 100644
--- a/resources/libraries/robot/performance/performance_actions.robot
+++ b/resources/libraries/robot/performance/performance_actions.robot
@@ -95,6 +95,7 @@
 | | ... | ramp_up_rate=${ramp_up_rate}
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=vppctl_runtime.yaml
+| | ... | oload=${telemetry_oload} | export=${True}
 | | Stop traffic on tg
 
 | Additional Statistics Action For bpf-runtime
@@ -131,6 +132,7 @@
 | | ... | ramp_up_rate=${ramp_up_rate}
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=bpf_runtime.yaml
+| | ... | oload=${telemetry_oload} | export=${False}
 | | Stop traffic on tg
 
 | Additional Statistics Action For perf-stat-runtime
@@ -167,6 +169,7 @@
 | | ... | ramp_up_rate=${ramp_up_rate}
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=perf_stat_runtime.yaml
+| | ... | oload=${telemetry_oload} | export=${False}
 | | Stop traffic on tg
 
 | Additional Statistics Action For vpp-runtime-iperf3
@@ -192,6 +195,7 @@
 | | | ... | affinity=${iperf_client_affinity}
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=vppctl_runtime.yaml
+| | ... | oload=${telemetry_oload} | export=${True}
 | | iPerf Client Stop Remote Exec | ${nodes['${iperf_client_node}']} | ${pids}
 
 | Additional Statistics Action For noop
@@ -206,6 +210,7 @@
 | |
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=vppctl_clear_stats.yaml
+| | ... | export=${False}
 
 | Additional Statistics Action For vpp-enable-packettrace
 | | [Documentation]
@@ -227,3 +232,4 @@
 | |
 | | Run Telemetry On All DUTs
 | | ... | ${nodes} | profile=vppctl_show_stats.yaml
+| | ... | export=${False}
diff --git a/resources/libraries/robot/performance/performance_utils.robot b/resources/libraries/robot/performance/performance_utils.robot
index 47c76661cb..1eb6710eff 100644
--- a/resources/libraries/robot/performance/performance_utils.robot
+++ b/resources/libraries/robot/performance/performance_utils.robot
@@ -113,6 +113,7 @@
 | | ... | ${average} | ${stdev}
 | | Set Test Variable | \${rate for teardown} | ${lower}
 | | # Stats at the discovered critical rate.
+| | Set Test Variable | ${telemetry_oload} | plr
 | | Send traffic at specified rate
 | | ... | rate=${lower}
 | | ... | trial_duration=${1.0}
@@ -196,12 +197,14 @@
 | | ${pdr} = | Set Variable | ${result[1].measured_low.target_tr}
 | | ${ndr} = | Set Variable | ${result[0].measured_low.target_tr}
 | | # We expect NDR and PDR to have different-looking stats.
+| | Set Test Variable | ${telemetry_oload} | pdr
 | | Send traffic at specified rate
 | | ... | rate=${pdr}
 | | ... | trial_duration=${1.0}
 | | ... | trial_multiplicity=${1}
 | | ... | use_latency=${use_latency}
 | | ... | duration_limit=${1.0}
+| | Set Test Variable | ${telemetry_oload} | ndr
 | | Run Keyword If | ${ndr} != ${pdr}
 | | ... | Send traffic at specified rate
 | | ... | rate=${ndr}
@@ -627,7 +630,6 @@
 | | ... | transaction_scale=${transaction_scale}
 | | ... | transaction_type=${transaction_type}
 | | ... | use_latency=${use_latency}
-| | # TODO: Ramp-up?
 
 | Stop Running Traffic
 | | [Documentation]
@@ -672,6 +674,7 @@
 | | ${unit} = | Set Variable If | """_cps""" in """${transaction_type}"""
 | | ... | cps | pps
 | | # The following also sets \${rate_for_teardown}
+| | Set Test Variable | ${telemetry_oload} | mrr
 | | ${results} = | Send traffic at specified rate
 | | ... | rate=${max_rate}
 | | ... | trial_duration=${trial_duration}