squidient 3.0.4__tar.gz → 3.0.7__tar.gz

{squidient-3.0.4 → squidient-3.0.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: squidient
-Version: 3.0.4
+Version: 3.0.7
 Summary: CI/CD framework for alya
 Project-URL: Homepage, https://alya.gitlab.bsc.es/alya/ci-cd/squidient
 Project-URL: Bug Tracker, https://alya.gitlab.bsc.es/alya/ci-cd/squidient/issues

{squidient-3.0.4 → squidient-3.0.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "squidient"
-version = "3.0.4"
+version = "3.0.7"
 authors = [
   { name="Damien Dosimont", email="damien.dosimont@gmail.com" },
 ]

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/cli/monitoring.py

@@ -82,7 +82,7 @@ def _run_cli(argv: Sequence[str]) -> int:
         monitoring.init()
 
         if "validate" == argv[0]:
-            monitoring.validate()
+            return 0 if monitoring.validate() else 1
 
         elif "build" == argv[0]:
             monitoring.lock(terraform)

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/cli/staging.py

@@ -137,7 +137,7 @@ def _run_cli(argv: Sequence[str]) -> int:
             cicd.init()
 
             if "validate" == argv[0]:
-                cicd.validate()
+                return 0 if cicd.validate() else 1
 
             elif "report" == argv[0]:
                 if len(argv) >= 2:

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/configure.py

@@ -56,6 +56,27 @@ from pathlib import Path
 # Use main(argv) to run it programmatically.
 
 
+def normalize_benchmark_failure_sender(sender):
+    if sender is None or sender == "":
+        return benchmark_failure_sender
+    return sender
+
+
+def normalize_benchmark_failure_copy(recipients):
+    if recipients is None:
+        return list(benchmark_failure_copy)
+    if isinstance(recipients, list):
+        candidates = recipients
+    else:
+        candidates = str(recipients).replace(";", ",").split(",")
+    normalized = []
+    for recipient in candidates:
+        recipient = str(recipient).strip()
+        if "@" in recipient and recipient not in normalized:
+            normalized.append(recipient)
+    return normalized
+
+
 def save_field(field, value):
     config = open_json(config_file)
     value = clean_value(value)
@@ -330,7 +351,8 @@ def generate_json():
 
 def gitlab_config(pipeline, directory, alya_ref, alya_branch, flex=False, system="bsc", platform="gpp", hosts=None, install=False,
                   build_selection_type=None, api=None, reservation=False, project=None, alamak=False, alamak_branch=None,
-                  alamak_revision=None, notify_failures=False):
+                  alamak_revision=None, notify_failures=False, failure_notification_sender=None,
+                  failure_notification_recipients=None):
 
     benchmark_presets = BuildPresets("benchmarks").get_presets()
     build_presets = BuildPresets("builds").get_presets()
@@ -383,8 +405,8 @@ def gitlab_config(pipeline, directory, alya_ref, alya_branch, flex=False, system
     console.section("Configuring flex")
     config["flex"] = flex
     config["benchmarks"]["notifyFailures"] = notify_failures
-    config["benchmarks"]["failureSender"] = benchmark_failure_sender
-    config["benchmarks"]["failureCopy"] = benchmark_failure_copy
+    config["benchmarks"]["failureSender"] = normalize_benchmark_failure_sender(failure_notification_sender)
+    config["benchmarks"]["failureCopy"] = normalize_benchmark_failure_copy(failure_notification_recipients)
     console.section("Configuring clean log")
     config["cleanLog"] = False
     console.section("Configuring system")
@@ -587,6 +609,8 @@ def _run_cli(argv: Sequence[str]) -> int:
         alamak_branch = None
         alamak_revision = None
         notify_failures = False
+        failure_notification_sender = None
+        failure_notification_recipients = None
 
         for arg in argv[0:]:
             if "--pipeline:" in arg:
@@ -637,6 +661,12 @@ def _run_cli(argv: Sequence[str]) -> int:
             if "--notify-failures" in arg:
                 notify_failures = True
                 console.step("Benchmark failure notifications enabled")
+            if "--failure-notification-sender:" in arg:
+                failure_notification_sender = arg.replace("--failure-notification-sender:", "")
+                console.step("Failure notification sender " + failure_notification_sender)
+            if "--failure-notification-recipients:" in arg:
+                failure_notification_recipients = arg.replace("--failure-notification-recipients:", "")
+                console.step("Failure notification recipients configured")
 
         gitlab_config(
             pipeline=pipeline,
@@ -656,6 +686,8 @@ def _run_cli(argv: Sequence[str]) -> int:
             alamak_branch=alamak_branch,
             alamak_revision=alamak_revision,
             notify_failures=notify_failures,
+            failure_notification_sender=failure_notification_sender,
+            failure_notification_recipients=failure_notification_recipients,
         )
         return 0
 

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/data/web/btest.html

@@ -37,6 +37,7 @@
         <th>Job</th>
         <th>Sbatch</th>
         <th>Output</th>
+        <th>Binding</th>
         <th>Alya</th>
         <th>Postprocess</th>
         <th>Job Id</th>
@@ -83,6 +84,9 @@
           t[x].output_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/test.out", null)
           t[x].output_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/test.err", null)
           t[x].output=t[x].output_out + " " + t[x].output_err
+          t[x].binding_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/binding.out", null)
+          t[x].binding_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/binding.err", null)
+          t[x].binding=t[x].binding_out + " " + t[x].binding_err
           t[x].alya_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/alya.out", null)
           t[x].alya_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/alya.err", null)
           t[x].alya=t[x].alya_out + " " + t[x].alya_err
@@ -105,7 +109,7 @@
               t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_KO, dir + "/alya.err", null, 2, "Job error")
             }
             else if (t[x].error=="binding"){
-              t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_BINDING, dir + "/alya.err", null, 3, "Binding error")
+              t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_BINDING, dir + "/binding.err", null, 3, "Binding error")
             }
             else if (t[x].error=="overload"){
               t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_WEIGHT, dir + "/alya.err", null, 4, "Overload error")

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/data/web/test.html

@@ -37,6 +37,7 @@
         <th>Job</th>
         <th>Sbatch</th>
         <th>Output</th>
+        <th>Binding</th>
         <th>Alya</th>
         <th>Postprocess</th>
         <th>Job Id</th>
@@ -84,6 +85,9 @@
           t[x].output_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/test.out", null)
           t[x].output_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/test.err", null)
           t[x].output=t[x].output_out + " " + t[x].output_err
+          t[x].binding_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/binding.out", null)
+          t[x].binding_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/binding.err", null)
+          t[x].binding=t[x].binding_out + " " + t[x].binding_err
           t[x].alya_out=testsuiteJSON.setUrl(testsuiteJSON.FILE_OK, dir + "/alya.out", null)
           t[x].alya_err=testsuiteJSON.setUrl(testsuiteJSON.FILE_KO, dir + "/alya.err", null)
           t[x].alya=t[x].alya_out + " " + t[x].alya_err
@@ -106,7 +110,7 @@
               t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_KO, dir + "/alya.err", null, 2, "Job error")
             }
             else if (t[x].error=="binding"){
-              t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_BINDING, dir + "/alya.err", null, 3, "Binding error")
+              t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_BINDING, dir + "/binding.err", null, 3, "Binding error")
             }
             else if (t[x].error=="overload"){
               t[x].status=testsuiteJSON.setUrl(testsuiteJSON.SYS_WEIGHT, dir + "/alya.err", null, 4, "Overload error")

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/monitoring/monitoring.py

@@ -139,7 +139,7 @@ class Monitoring(Staging):
             console.error(message + "\nDetails written to " + log_path, level=1)
         if len(failures) > 0:
             raise WorkflowFailure("Benchmark notification email could not be sent")
-        super().validate()
+        return super().validate()
 
     def write_notification_failure_log(self, failures):
         log_dir = os.path.dirname(benchmark_notification_failure_log)
@@ -197,8 +197,6 @@ class Monitoring(Staging):
             console.report("web/monitoring_test.html", level=1)
             print_line()
             self.clean_locks()
-            if not status:
-                raise WorkflowFailure("Monitoring test phase failed")
 
     def prepare_monitoring_test_builds(self, fake=False):
         if not fake:

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/staging/staging.py

@@ -327,8 +327,6 @@ class Staging:
             console.report("web/staging_test.html", level=1)
             print_line()
             self.clean_locks()
-            if not status:
-                raise WorkflowFailure("Test phase failed")
 
     def create_test_orchestrator(self):
         return TestOrchestrator(
@@ -568,10 +566,12 @@ class Staging:
         self.clean_locks()
 
     def validate(self):
+        console.section("Validation result")
         if self._report.get_valid():
+            console.success("Report is valid", level=1)
             return True
-        else:
-            raise WorkflowFailure("Report is not valid")
+        console.error("Report is not valid", level=1)
+        return False
 
     def report_push(self):
         with self.timer("staging.report_push", local=False, system="bsc"):

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/testing/benchmarks/failure_notifier.py

@@ -20,6 +20,7 @@
 from email.message import EmailMessage
 from ...definitions import benchmark_failure_log_files
 import os
+import re
 import shutil
 import smtplib
 import subprocess
@@ -68,9 +69,24 @@ class BenchmarkFailureNotifier:
         self._from_address = from_address
         if copy_recipients is None:
             copy_recipients = []
-        self._copy_recipients = copy_recipients
+        self._copy_recipients = self.normalize_recipients(copy_recipients)
         self._mail_sender = mail_sender or self.send_mail
 
+    @staticmethod
+    def normalize_recipients(recipients):
+        if recipients is None:
+            return []
+        if isinstance(recipients, str):
+            candidates = re.split(r"[,;]", recipients)
+        else:
+            candidates = recipients
+        normalized = []
+        for candidate in candidates:
+            candidate = str(candidate).strip()
+            if "@" in candidate and candidate not in normalized:
+                normalized.append(candidate)
+        return normalized
+
     def notify(self, report):
         recipients = self.recipients(report.get("authors", []))
         if len(recipients) == 0:

{squidient-3.0.4 → squidient-3.0.7}/src/squidient/testing/cases/instance.py

@@ -46,6 +46,8 @@ from ...utils.utils import (
 
 SBATCH_OUTPUT_FILE = "sbatch.out"
 SBATCH_ERROR_FILE = "sbatch.err"
+BINDING_OUTPUT_FILE = "binding.out"
+BINDING_ERROR_FILE = "binding.err"
 
 
 def file_contains(path, text):
@@ -125,6 +127,7 @@ class Instance:
         self._report = {}
 
         self._alya_cmd = ""
+        self._binding_cmd = ""
         self._pp_cmd = ""
         self._exec_modules = []
 
@@ -241,10 +244,13 @@ class Instance:
         alya_custom = ""
         postprocess_custom = ""
         alya_out = " >> alya.out 2>> alya.err"
+        binding_out = " >> " + BINDING_OUTPUT_FILE + " 2>> " + BINDING_ERROR_FILE
         run_exec = "mpirun"
         run_proc = "-np"
         run_sep = ":"
-        alya_default = run_exec + " " + self._run_bind + " " + self._run_map + " " + self._run_options + " $" + self._binding_report + " " + run_proc + " " + self._mpi + " $ALYA " + self._name
+        run_prefix = run_exec + " " + self._run_bind + " " + self._run_map + " " + self._run_options
+        alya_default = run_prefix + " " + run_proc + " " + self._mpi + " $ALYA " + self._name
+        self._binding_cmd = 'if [ -n "$' + self._binding_report + '" ]; then ' + run_prefix + " $" + self._binding_report + " " + run_proc + " " + self._mpi + " /bin/true" + binding_out + "; fi"
         postprocess_output_redirect = " >> post.out 2>> post.err"
         postprocess_default = 'echo "Nothing to do"'
         postprocess_command = ""
@@ -272,7 +278,7 @@ class Instance:
             if "postprocess" in self._execution["commands"]:
                 postprocess_custom = self._execution["commands"]["postprocess"]
         if alya_custom != "":
-            alya_custom = alya_custom.replace("[RUN]", run_exec + " " + self._run_bind + " " + self._run_map + " " + self._run_options)
+            alya_custom = alya_custom.replace("[RUN]", run_prefix)
             alya_custom = alya_custom.replace("[NP]", run_proc)
             alya_custom = alya_custom.replace("[MPI]", self._mpi)
             alya_custom = alya_custom.replace("[NAME]", self._name)
@@ -367,8 +373,8 @@ class Instance:
             self._script += pre_exec + "\n"
         self._script += 'if mpirun -V | grep "Open MPI"; then export ' + self._binding_report + '="--report-bindings"' + "; fi\n"
         self._script += 'if mpirun -V | grep "Intel"; then export ' + self._binding_report + '="-print-rank-map"' + "; fi\n"
-        self._script += "echo $" + self._binding_report + "\n"
-        self._script += "rm -f alya.out alya.err infiniband.out\n"
+        self._script += "rm -f alya.out alya.err " + BINDING_OUTPUT_FILE + " " + BINDING_ERROR_FILE + " infiniband.out\n"
+        self._script += self._binding_cmd + "\n"
         self._script += "cat /sys/class/infiniband/hfi1_0/ports/1/counters/port_rcv_data > infiniband.out\n"
         self._script += "cat /sys/class/infiniband/hfi1_0/ports/1/counters/port_xmit_data >> infiniband.out\n"
         self._script += "date '+%s' >> time.out\n"
@@ -491,7 +497,10 @@ class Instance:
                 self.retry()
             elif file_contains(self._report_path + "/alya.err", "All nodes which are allocated for this job are already filled"):
                 self._report["error"] = "overload"
-            elif file_contains(self._report_path + "/alya.err", "A request was made to bind to that would result in binding more"):
+            elif (
+                file_contains(self._report_path + "/" + BINDING_ERROR_FILE, "A request was made to bind to that would result in binding more")
+                or file_contains(self._report_path + "/alya.err", "A request was made to bind to that would result in binding more")
+            ):
                 self._report["error"] = "binding"
             elif file_contains(self._report_path + "/alya.err", "Integer overflow in xmallocarray"):
                 self.retry()
@@ -646,6 +655,8 @@ class Instance:
         open(self._report_path + "/test.err", 'a+').close()
         open(self._report_path + "/alya.out", 'a+').close()
         open(self._report_path + "/alya.err", 'a+').close()
+        open(self._report_path + "/" + BINDING_OUTPUT_FILE, 'a+').close()
+        open(self._report_path + "/" + BINDING_ERROR_FILE, 'a+').close()
         open(self._report_path + "/post.out", 'a+').close()
         open(self._report_path + "/post.err", 'a+').close()
         open(self._report_path + "/" + SBATCH_OUTPUT_FILE, 'a+').close()
@@ -659,6 +670,8 @@ class Instance:
             "test.err",
             "alya.out",
             "alya.err",
+            BINDING_OUTPUT_FILE,
+            BINDING_ERROR_FILE,
             "post.out",
             "post.err",
             SBATCH_OUTPUT_FILE,