Merge pull request #31 from jameshcorbett/pals-apinfo-tests

cray-pals: add apinfo content tests

t/scripts/apinfo_checker.py

@@ -0,0 +1,90 @@
+#!/usr/bin/env python3
+
+import struct
+import os
+import json
+import sys
+
+
+def get_comm_profiles(apinfo, size, offset, count):
+    comms = []
+    for comm in _get_structs(apinfo, size, offset, count, "40siii"):
+        tokenid, vni, vlan, traffic_classes = comm
+        comms.append(
+            {
+                "tokenid": tokenid,
+                "vni": vni,
+                "vlan": vlan,
+                "traffic_classes": traffic_classes,
+            }
+        )
+    return comms
+
+
+def get_cmds(apinfo, size, offset, count):
+    cmds = []
+    for cmd in _get_structs(apinfo, size, offset, count, "iii"):
+        npes, pes_per_node, cpus_per_pe = cmd
+        cmds.append(
+            {"npes": npes, "pes_per_node": pes_per_node, "cpus_per_pe": cpus_per_pe}
+        )
+    return cmds
+
+
+def get_pes(apinfo, size, offset, count):
+    pes = []
+    for pe in _get_structs(apinfo, size, offset, count, "iii"):
+        localidx, cmdidx, nodeidx = pe
+        pes.append({"localidx": localidx, "cmdidx": cmdidx, "nodeidx": nodeidx})
+    return pes
+
+
+def get_nodes(apinfo, size, offset, count):
+    nodes = []
+    for node in _get_structs(apinfo, size, offset, count, "i64s"):
+        node_id, hostname = node
+        nodes.append(
+            {"id": node_id, "hostname": hostname.split(b"\x00")[0].decode("ascii")}
+        )
+    return nodes
+
+
+def get_nics(apinfo, size, offset, count):
+    nics = []
+    for pe in _get_structs(apinfo, size, offset, count, "ii40s"):
+        nodeidx, address_type, address = pe
+        nics.append(
+            {
+                "nodeidx": nodeidx,
+                "address_type": address_type,
+                "address": address.split(b"\x00")[0].decode("ascii"),
+            }
+        )
+    return nics
+
+
+def _get_structs(apinfo, size, offset, count, format):
+    for i in range(count):
+        yield struct.unpack(format, apinfo[offset : offset + size])
+        offset += size
+
+
+def main():
+    with open(os.environ["PALS_APINFO"], "rb") as fd:
+        apinfo = fd.read()
+    header_format = "iNNNiNNiNNiNNiNNi"
+    header = struct.unpack(header_format, apinfo[: struct.calcsize(header_format)])
+    apinfo_dict = {
+        "version": header[0],
+        "comm_profiles": get_comm_profiles(apinfo, header[2], header[3], header[4]),
+        "cmds": get_cmds(apinfo, header[5], header[6], header[7]),
+        "pes": get_pes(apinfo, header[8], header[9], header[10]),
+        "nodes": get_nodes(apinfo, header[11], header[12], header[13]),
+        "nics": get_nics(apinfo, header[14], header[15], header[16]),
+    }
+    json.dump(apinfo_dict, sys.stdout)
+    print()
+
+
+if __name__ == "__main__":
+    main()

t/t1001-cray-pals.t

@@ -79,6 +79,42 @@ test_expect_success 'shell: pals shell plugin creates apinfo file' '
 	&& test ! -z \$PALS_APINFO && test -f \$PALS_APINFO"
 '
 
+test_expect_success HAVE_JQ 'shell: apinfo file contents are valid for one task' '
+	apinfo=$(flux mini run -o userrc=$(pwd)/$USERRC_NAME -N1 -n1 ${PYTHON:-python3} \
+	${SHARNESS_TEST_SRCDIR}/scripts/apinfo_checker.py) &&
+	echo "$apinfo" | jq -e ".version == 1" &&
+	echo "$apinfo" | jq -e ".cmds[0].npes == 1" &&
+	echo "$apinfo" | jq -e ".pes[0].localidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[0].cmdidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[0].nodeidx == 0" &&
+	echo "$apinfo" | jq -e ".nodes[0].id == 0" &&
+	test $(hostname) = $(echo "$apinfo" | jq -r .nodes[0].hostname) &&
+	echo "$apinfo" | jq ".nics | length == 0" &&
+	echo "$apinfo" | jq ".comm_profiles | length == 0" &&
+	echo "$apinfo" | jq -e ".nodes | length == 1" &&
+	echo "$apinfo" | jq -e ".cmds | length == 1" &&
+	echo "$apinfo" | jq -e ".pes | length == 1"
+'
+
+test_expect_success HAVE_JQ 'shell: apinfo file contents are valid for multiple tasks' '
+	apinfo=$(flux mini run -o userrc=$(pwd)/$USERRC_NAME -N1 -n2 --label-io \
+	${PYTHON:-python3} ${SHARNESS_TEST_SRCDIR}/scripts/apinfo_checker.py \
+	| sed -n "s/^1: //p") &&
+	echo "$apinfo" | jq -e ".cmds[0].npes == 2" &&
+	echo "$apinfo" | jq -e ".cmds[0].pes_per_node == 2" &&
+	echo "$apinfo" | jq -e ".pes[0].localidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[1].localidx == 1" &&
+	echo "$apinfo" | jq -e ".pes[0].cmdidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[1].cmdidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[0].nodeidx == 0" &&
+	echo "$apinfo" | jq -e ".pes[1].nodeidx == 0" &&
+	echo "$apinfo" | jq -e ".nics | length == 0" &&
+	echo "$apinfo" | jq -e ".comm_profiles | length == 0" &&
+	echo "$apinfo" | jq -e ".nodes | length ==1" &&
+	echo "$apinfo" | jq -e ".cmds | length == 1" &&
+	echo "$apinfo" | jq -e ".pes | length == 2"
+'
+
 test_expect_success 'shell: pals shell plugin handles resource oversubscription' '
 	flux mini run -o userrc=$(pwd)/$USERRC_NAME -N1 -n2 -oper-resource.type=core -oper-resource.count=2 \
 	env | grep PALS_RANKID=3 &&