Github user NicoK commented on a diff in the pull request:
https://github.com/apache/flink/pull/3721#discussion_r112248789
--- Diff: flink-dist/src/main/flink-bin/bin/config.sh ---
@@ -398,3 +428,106 @@ readSlaves() {
useOffHeapMemory() {
[[ "`echo ${FLINK_TM_OFFHEAP} | tr '[:upper:]' '[:lower:]'`" == "true"
]]
}
+
+HAVE_AWK=
+# same as
org.apache.flink.runtime.taskexecutor.TaskManagerServices.calculateNetworkBuf(long
totalJavaMemorySize, Configuration config)
+calculateNetworkBuf() {
+ local network_buffers_bytes
+ if [ "${FLINK_TM_HEAP}" -le "0" ]; then
+ echo "Variable 'FLINK_TM_HEAP' not set (usually read from
'${KEY_TASKM_MEM_SIZE}' in ${FLINK_CONF_FILE})."
+ exit 1
+ fi
+
+ if [[ "${FLINK_TM_NET_BUF_MIN}" = "${FLINK_TM_NET_BUF_MAX}" ]]; then
+ # fix memory size for network buffers
+ network_buffers_bytes=${FLINK_TM_NET_BUF_MIN}
+ else
+ if [[ "${FLINK_TM_NET_BUF_MIN}" -gt "${FLINK_TM_NET_BUF_MAX}" ]];
then
+ echo "[ERROR] Configured TaskManager network buffer memory
min/max '${FLINK_TM_NET_BUF_MIN}'/'${FLINK_TM_NET_BUF_MAX}' are not valid."
+ echo "Min must be less than or equal to max."
+ echo "Please set '${KEY_TASKM_NET_BUF_MIN}' and
'${KEY_TASKM_NET_BUF_MAX}' in ${FLINK_CONF_FILE}."
+ exit 1
+ fi
+
+ # Bash only performs integer arithmetic so floating point
computation is performed using awk
+ if [[ -z "${HAVE_AWK}" ]] ; then
+ command -v awk >/dev/null 2>&1
+ if [[ $? -ne 0 ]]; then
+ echo "[ERROR] Program 'awk' not found."
+ echo "Please install 'awk' or define
'${KEY_TASKM_NET_BUF_MIN}' and '${KEY_TASKM_NET_BUF_MAX}' instead of
'${KEY_TASKM_NET_BUF_FRACTION}' in ${FLINK_CONF_FILE}."
+ exit 1
+ fi
+ HAVE_AWK=true
+ fi
+
+ # We calculate the memory using a fraction of the total memory
+ if [[ `awk '{ if ($1 > 0.0 && $1 < 1.0) print "1"; }' <<<
"${FLINK_TM_NET_BUF_FRACTION}"` != "1" ]]; then
+ echo "[ERROR] Configured TaskManager network buffer memory
fraction '${FLINK_TM_NET_BUF_FRACTION}' is not a valid value."
+ echo "It must be between 0.0 and 1.0."
+ echo "Please set '${KEY_TASKM_NET_BUF_FRACTION}' in
${FLINK_CONF_FILE}."
+ exit 1
+ fi
+
+ network_buffers_bytes=`awk "BEGIN { x =
lshift(${FLINK_TM_HEAP},20) * ${FLINK_TM_NET_BUF_FRACTION}; netbuf = x >
${FLINK_TM_NET_BUF_MAX} ? ${FLINK_TM_NET_BUF_MAX} : x < ${FLINK_TM_NET_BUF_MIN}
? ${FLINK_TM_NET_BUF_MIN} : x; printf \"%.0f\n\", netbuf }"`
+ fi
+
+ # recalculate the JVM heap memory by taking the network buffers into
account
--- End diff --
no, actually, the user may give the `FLINK_TM_HEAP` environment variable or
configure the "flink heap size" via `taskmanager.heap.mb` but this is not the
real "heap" size - rather the overall memory size used by flink (including
off-heap). So this function removes the off-heap part and returns the real heap
sizes to use with `-Xmx` and `-Xms`
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
