This is an automated email from the ASF dual-hosted git repository.
laiyingchun pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-pegasus.git
The following commit(s) were added to refs/heads/master by this push:
new 21d29604e fix(hdfs): export LIBHDFS_OPTS=-Xrs to ensure server could
exit normally (#1911)
21d29604e is described below
commit 21d29604e9e79ac8b3b596bbf6f7641e69f62d6f
Author: Yingchun Lai <[email protected]>
AuthorDate: Wed Feb 21 15:25:40 2024 +0800
fix(hdfs): export LIBHDFS_OPTS=-Xrs to ensure server could exit normally
(#1911)
Resolve https://github.com/apache/incubator-pegasus/issues/1913.
Reduces the use of operating system signals by the JVM by exporting
LIBHDFS_OPTS="-Xrs" to resolve the issue of unexpected exit order of
JVM threads and Pegasus threads.
See details:
https://docs.oracle.com/javase/8/docs/technotes/tools/unix/java.html
---
scripts/config_hdfs.sh | 5 ++++
src/block_service/hdfs/hdfs_service.cpp | 52 ++++++++++++++++++++++++++-------
2 files changed, 47 insertions(+), 10 deletions(-)
diff --git a/scripts/config_hdfs.sh b/scripts/config_hdfs.sh
index 0ad1260c6..de8da9fbf 100755
--- a/scripts/config_hdfs.sh
+++ b/scripts/config_hdfs.sh
@@ -58,6 +58,11 @@ if [ ! -d "$HADOOP_HOME/etc/hadoop" ] || [ ! -d
"$HADOOP_HOME/share/hadoop" ]; t
echo "HADOOP_HOME must be set to the location of your Hadoop jars and
core-site.xml."
return 1
fi
+
+# Reduces the use of operating system signals by the JVM. See details:
+# https://docs.oracle.com/javase/8/docs/technotes/tools/unix/java.html
+export LIBHDFS_OPTS="-Xrs"
+
export CLASSPATH=$CLASSPATH:$HADOOP_HOME/etc/hadoop/
for f in $HADOOP_HOME/share/hadoop/common/lib/*.jar; do
export CLASSPATH=$CLASSPATH:$f
diff --git a/src/block_service/hdfs/hdfs_service.cpp
b/src/block_service/hdfs/hdfs_service.cpp
index b8e9dee4e..e30371049 100644
--- a/src/block_service/hdfs/hdfs_service.cpp
+++ b/src/block_service/hdfs/hdfs_service.cpp
@@ -18,10 +18,13 @@
#include <errno.h>
#include <fcntl.h>
#include <rocksdb/env.h>
+#include <stdlib.h>
#include <algorithm>
#include <type_traits>
+#include <unordered_set>
#include <utility>
+#include "fmt/core.h"
#include "hdfs/hdfs.h"
#include "hdfs_service.h"
#include "rocksdb/slice.h"
@@ -69,28 +72,57 @@ namespace block_service {
DEFINE_TASK_CODE(LPC_HDFS_SERVICE_CALL, TASK_PRIORITY_COMMON,
THREAD_POOL_BLOCK_SERVICE)
+namespace {
+static const char *LIBHDFS_OPTS = "LIBHDFS_OPTS";
+static const char *REDUCE_SIGNAL_OPT = "-Xrs";
+
+bool check_LIBHDFS_OPTS()
+{
+ char *str = ::getenv(LIBHDFS_OPTS);
+ if (str == nullptr) {
+ return false;
+ }
+
+ std::unordered_set<std::string> opts;
+ utils::split_args(str, opts, ' ');
+ return opts.count(REDUCE_SIGNAL_OPT) > 0;
+}
+
+void adjust_LIBHDFS_OPTS()
+{
+ if (!check_LIBHDFS_OPTS()) {
+ // Reduces the use of operating system signals by the JVM. See details:
+ // https://docs.oracle.com/javase/8/docs/technotes/tools/unix/java.html
+ std::string new_env;
+ char *old_env = ::getenv(LIBHDFS_OPTS);
+ if (old_env == nullptr) {
+ new_env = REDUCE_SIGNAL_OPT;
+ } else {
+ new_env = fmt::format("{} {}", old_env, REDUCE_SIGNAL_OPT);
+ }
+ ::setenv(LIBHDFS_OPTS, new_env.c_str(), 1);
+ LOG_WARNING("update '{}' to '{}'", LIBHDFS_OPTS, new_env);
+ }
+ CHECK_TRUE(check_LIBHDFS_OPTS());
+}
+} // anonymous namespace
+
hdfs_service::hdfs_service()
{
_read_token_bucket.reset(new folly::DynamicTokenBucket());
_write_token_bucket.reset(new folly::DynamicTokenBucket());
}
-hdfs_service::~hdfs_service()
-{
- // We should not call hdfsDisconnect() here if jvm has exited.
- // And there is no simple, safe way to call hdfsDisconnect()
- // when process terminates (the proper solution is likely to create a
- // signal handler to detect when the process is killed, but we would still
- // leak when pegasus crashes).
- //
- // close();
-}
+hdfs_service::~hdfs_service() { close(); }
error_code hdfs_service::initialize(const std::vector<std::string> &args)
{
if (args.size() < 1) {
return ERR_INVALID_PARAMETERS;
}
+
+ adjust_LIBHDFS_OPTS();
+
// Name_node and root_path should be set in args of block_service
configuration.
// If no path was configured, just use "/" as default root path.
_hdfs_name_node = args[0];
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
