Skip to contentSkip to conversation

ci: catch coredumps

RFC PATCH Superseded David Marchand 3 participants 6 messages Jan 11, 2021 – Mar 3, 2021

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 ci: catch coredumps

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.

In some situations (like explicitly call rte_panic), coredump generation must be disabled to avoid false positives.

---
Sending as a RFC, as this is a "nice to have" patch I had in store for some time, but I did not see the actual need so far.

We could attach the coredumps in GHA result, but it would be hardly usable without attaching all generated binaries...
Opinions?

---
 .ci/linux-build.sh    |  8 ++++++++
 app/test/test_debug.c | 11 +++++++++--
 app/test/test_mbuf.c  |  9 ++++++++-
 3 files changed, 25 insertions(+), 3 deletions(-)
.ci/linux-build.sh +8 -0
index d2c821adf3..d00a5804b4 100755
@@ -57,7 +57,11 @@ meson build --werror $OPTS
 ninja -C build
 
 if [ "$AARCH64" != "true" ]; then
+    ulimit -c unlimited
+    sudo sysctl -w kernel.core_pattern=/tmp/dpdk-core.%e.%p
+
     devtools/test-null.sh
+    ! ls /tmp/dpdk-core.*.* 2>/dev/null
 fi
 
 if [ "$ABI_CHECKS" = "true" ]; then
@@ -102,5 +106,9 @@ if [ "$ABI_CHECKS" = "true" ]; then
 fi
 
 if [ "$RUN_TESTS" = "true" ]; then
+    ulimit -c unlimited
+    sudo sysctl -w kernel.core_pattern=/tmp/dpdk-core.%e.%p
+
     sudo meson test -C build --suite fast-tests -t 3
+    ! ls /tmp/dpdk-core.*.* 2>/dev/null
 fi
app/test/test_debug.c +9 -2
index 834a7386f5..23b24db177 100644
@@ -4,6 +4,8 @@
 
 #include <stdio.h>
 #include <stdint.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 #include <unistd.h>
 
@@ -28,9 +30,14 @@ test_panic(void)
 
 	pid = fork();
 
-	if (pid == 0)
+	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_panic("Test Debug\n");
-	else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
app/test/test_mbuf.c +8 -1
index a40f7d4883..47a7b197d7 100644
@@ -1174,6 +1174,8 @@ test_refcnt_mbuf(void)
 }
 
 #include <unistd.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 
 /* use fork() to test mbuf errors panic */
@@ -1186,9 +1188,14 @@ verify_mbuf_check_panics(struct rte_mbuf *buf)
 	pid = fork();
 
 	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_mbuf_sanity_check(buf, 1); /* should panic */
 		exit(0);  /* return normally if it doesn't panic */
-	} else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
Signature · 1 line
2.23.0
#1 lore raw
Quoted 8 lines · David Marchand

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.

In some situations (like explicitly call rte_panic), coredump generation must be disabled to avoid false positives.

Quoted 8 lines

---
Sending as a RFC, as this is a "nice to have" patch I had in store for some time, but I did not see the actual need so far.

We could attach the coredumps in GHA result, but it would be hardly usable without attaching all generated binaries...
Opinions?

I think it's a good start. We may not need to attach the binaries at all - if we have access to gdb, we can run a script to just run simple commands like 'thread apply all bt' and maybe some info commands. That can all be stuffed into the logs. Usually a source level backtrace is enough to work through what happened.

Quoted 91 lines
---
 .ci/linux-build.sh    |  8 ++++++++
 app/test/test_debug.c | 11 +++++++++--
 app/test/test_mbuf.c  |  9 ++++++++-
 3 files changed, 25 insertions(+), 3 deletions(-)
.ci/linux-build.sh +8 -0
index d2c821adf3..d00a5804b4 100755
@@ -57,7 +57,11 @@ meson build --werror $OPTS
 ninja -C build
 
 if [ "$AARCH64" != "true" ]; then
+    ulimit -c unlimited
+    sudo sysctl -w kernel.core_pattern=/tmp/dpdk-core.%e.%p
+
     devtools/test-null.sh
+    ! ls /tmp/dpdk-core.*.* 2>/dev/null
 fi
 
 if [ "$ABI_CHECKS" = "true" ]; then
@@ -102,5 +106,9 @@ if [ "$ABI_CHECKS" = "true" ]; then
 fi
 
 if [ "$RUN_TESTS" = "true" ]; then
+    ulimit -c unlimited
+    sudo sysctl -w kernel.core_pattern=/tmp/dpdk-core.%e.%p
+
     sudo meson test -C build --suite fast-tests -t 3
+    ! ls /tmp/dpdk-core.*.* 2>/dev/null
 fi
app/test/test_debug.c +9 -2
index 834a7386f5..23b24db177 100644
@@ -4,6 +4,8 @@
 
 #include <stdio.h>
 #include <stdint.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 #include <unistd.h>
 
@@ -28,9 +30,14 @@ test_panic(void)
 
 	pid = fork();
 
-	if (pid == 0)
+	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_panic("Test Debug\n");
-	else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
app/test/test_mbuf.c +8 -1
index a40f7d4883..47a7b197d7 100644
@@ -1174,6 +1174,8 @@ test_refcnt_mbuf(void)
 }
 
 #include <unistd.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 
 /* use fork() to test mbuf errors panic */
@@ -1186,9 +1188,14 @@ verify_mbuf_check_panics(struct rte_mbuf *buf)
 	pid = fork();
 
 	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_mbuf_sanity_check(buf, 1); /* should panic */
 		exit(0);  /* return normally if it doesn't panic */
-	} else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
#2 lore raw

PATCH v2 1/1 ci: catch coredumps

In reply to David Marchand

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.
When unit tests fail, it can also help checking for coredumps as it could give more insights on what happened.

In some situations (like explicit call to rte_panic), coredump generation must be disabled to avoid false positives.

---
Changelog since RFC v1
- removed RFC,
- pushed coredumps into gdb for in-situ analysis,
- gdb presence is used to enable the check. Travis config is left untouched for now,

---
 .ci/linux-build.sh          | 37 ++++++++++++++++++++++++++++++++++---
 .github/workflows/build.yml |  4 ++++
 app/test/test_debug.c       | 11 +++++++++--
 app/test/test_mbuf.c        |  9 ++++++++-
 4 files changed, 55 insertions(+), 6 deletions(-)
.ci/linux-build.sh +34 -3
index d2c821adf3..26c30a2301 100755
@@ -4,7 +4,10 @@ on_error() {
     if [ $? = 0 ]; then
         exit
     fi
-    FILES_TO_PRINT="build/meson-logs/testlog.txt build/.ninja_log build/meson-logs/meson-log.txt"
+    FILES_TO_PRINT="build/meson-logs/testlog.txt"
+    FILES_TO_PRINT="$FILES_TO_PRINT build/.ninja_log"
+    FILES_TO_PRINT="$FILES_TO_PRINT build/meson-logs/meson-log.txt"
+    FILES_TO_PRINT="$FILES_TO_PRINT build/gdb.log"
 
     for pr_file in $FILES_TO_PRINT; do
         if [ -e "$pr_file" ]; then
@@ -30,6 +33,26 @@ install_libabigail() {
     rm ${version}.tar.gz
 }
 
+configure_coredump() {
+    # No point in configuring coredump without gdb
+    which gdb >/dev/null || return 0
+    ulimit -c unlimited
+    sudo sysctl -w kernel.core_pattern=/tmp/dpdk-core.%e.%p
+}
+
+catch_coredump() {
+    ls /tmp/dpdk-core.*.* 2>/dev/null || return 0
+    for core in /tmp/dpdk-core.*.*; do
+        binary=$(sudo readelf -n $core |grep $(pwd)/build/ 2>/dev/null |head -n1)
+        [ -x $binary ] || binary=
+        sudo gdb $binary -c $core \
+            -ex 'info threads' \
+            -ex 'thread apply all bt full' \
+            -ex 'quit'
+    done |tee -a build/gdb.log
+    return 1
+}
+
 if [ "$AARCH64" = "true" ]; then
     # convert the arch specifier
     OPTS="$OPTS --cross-file config/arm/arm64_armv8_linux_gcc"
@@ -57,7 +80,11 @@ meson build --werror $OPTS
 ninja -C build
 
 if [ "$AARCH64" != "true" ]; then
-    devtools/test-null.sh
+    failed=
+    configure_coredump
+    devtools/test-null.sh || failed="true"
+    catch_coredump
+    [ "$failed" != "true" ]
 fi
 
 if [ "$ABI_CHECKS" = "true" ]; then
@@ -102,5 +129,9 @@ if [ "$ABI_CHECKS" = "true" ]; then
 fi
 
 if [ "$RUN_TESTS" = "true" ]; then
-    sudo meson test -C build --suite fast-tests -t 3
+    failed=
+    configure_coredump
+    sudo meson test -C build --suite fast-tests -t 3 || failed="true"
+    catch_coredump
+    [ "$failed" != "true" ]
 fi
.github/workflows/build.yml +4 -0
index a5b579adda..786525e5a3 100644
@@ -104,6 +104,9 @@ jobs:
       if: env.AARCH64 == 'true'
       run: sudo apt install -y gcc-aarch64-linux-gnu libc6-dev-arm64-cross
         pkg-config-aarch64-linux-gnu
+    - name: Install test tools packages
+      if: env.AARCH64 != 'true' || env.RUN_TESTS == 'true'
+      run: sudo apt install -y gdb
     - name: Install doc generation packages
       if: env.BUILD_DOCS == 'true'
       run: sudo apt install -y doxygen graphviz python3-sphinx
@@ -124,3 +127,4 @@ jobs:
           build/meson-logs/testlog.txt
           build/.ninja_log
           build/meson-logs/meson-log.txt
+          build/gdb.log
app/test/test_debug.c +9 -2
index 834a7386f5..23b24db177 100644
@@ -4,6 +4,8 @@
 
 #include <stdio.h>
 #include <stdint.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 #include <unistd.h>
 
@@ -28,9 +30,14 @@ test_panic(void)
 
 	pid = fork();
 
-	if (pid == 0)
+	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_panic("Test Debug\n");
-	else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
app/test/test_mbuf.c +8 -1
index a40f7d4883..47a7b197d7 100644
@@ -1174,6 +1174,8 @@ test_refcnt_mbuf(void)
 }
 
 #include <unistd.h>
+#include <sys/resource.h>
+#include <sys/time.h>
 #include <sys/wait.h>
 
 /* use fork() to test mbuf errors panic */
@@ -1186,9 +1188,14 @@ verify_mbuf_check_panics(struct rte_mbuf *buf)
 	pid = fork();
 
 	if (pid == 0) {
+		struct rlimit rl;
+
+		/* No need to generate a coredump when panicking. */
+		rl.rlim_cur = rl.rlim_max = 0;
+		setrlimit(RLIMIT_CORE, &rl);
 		rte_mbuf_sanity_check(buf, 1); /* should panic */
 		exit(0);  /* return normally if it doesn't panic */
-	} else if (pid < 0){
+	} else if (pid < 0) {
 		printf("Fork Failed\n");
 		return -1;
 	}
Signature · 1 line
2.23.0
Acked-by on v2 1/1
#3 lore raw
Quoted 18 lines · David Marchand

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.
When unit tests fail, it can also help checking for coredumps as it could give more insights on what happened.

In some situations (like explicit call to rte_panic), coredump generation must be disabled to avoid false positives.

---
Changelog since RFC v1
- removed RFC,
- pushed coredumps into gdb for in-situ analysis,
- gdb presence is used to enable the check. Travis config is left untouched for now,

Quoted 6 lines
---
 .ci/linux-build.sh          | 37 ++++++++++++++++++++++++++++++++++---
 .github/workflows/build.yml |  4 ++++
 app/test/test_debug.c       | 11 +++++++++--
 app/test/test_mbuf.c        |  9 ++++++++-
 4 files changed, 55 insertions(+), 6 deletions(-)

This fixes an issue for me, with running the debug_autotest under a reproducible build CI environment.

Signature · 2 lines
Kind regards,
Luca Boccassi
Acked-by on v2 1/1
#4 lore raw
In reply to David Marchand
Quoted 7 lines · David Marchand

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.
When unit tests fail, it can also help checking for coredumps as it could give more insights on what happened.

Quoted 5 lines

In some situations (like explicit call to rte_panic), coredump generation must be disabled to avoid false positives.

---

I had this in my backlog and accidentally marked it 'visited' - sorry about that.

author Acked-by Luca Boccassi Acked-by Aaron Conole on v2 1/1
#5 lore raw
In reply to David Marchand
Quoted 11 lines · David Marchand

Parts of the unit tests code rely on forked/secondary processes (expectedly) failing.
A crash in those situations could be missed so add a check on coredumps presence after unit tests have run.
When unit tests fail, it can also help checking for coredumps as it could give more insights on what happened.

In some situations (like explicit call to rte_panic), coredump generation must be disabled to avoid false positives.

As suggested by Luca, marking for backport.
Cc: [email protected]

Quoted 1 line
Signature · 1 line
David Marchand