Make memory tuning cgroups v2 compatible

The old implementation only checked the sysfs files corresponding to
cgroups v1. Most Linux distribution kernels have switched to cgroups
v2 several years ago, which uses a different set of paths for various
settings (memory limits included).

The net effect of the old implementation was that it completely
ignored the memory limits, if the container was run in a kernel with
cgroups v2 (in those kernels cgroups v1 are disabled by default).

docker-entrypoint-initdb.d/001_timescaledb_tune.sh

@@ -18,9 +18,16 @@ fi
 
 if [ -z "${TS_TUNE_MEMORY:-}" ]; then
     # See if we can get the container's total allocated memory from the cgroups metadata
-    if [ -f /sys/fs/cgroup/memory/memory.limit_in_bytes ]; then
+    # Try with cgroups v2 first.
+    if [ -f /sys/fs/cgroup/cgroup.controllers ]; then
+        TS_TUNE_MEMORY=$(cat /sys/fs/cgroup/memory.max)
+        TS_CGROUPS_MAX_MEM=true
+    # cgroups v2 is not available, try with cgroups v1
+    elif [ -f /sys/fs/cgroup/memory/memory.limit_in_bytes ]; then
         TS_TUNE_MEMORY=$(cat /sys/fs/cgroup/memory/memory.limit_in_bytes)
-
+        TS_CGROUPS_MAX_MEM=true
+    fi
+    if [ "${TS_CGROUPS_MAX_MEM:-false}" != "false" ]; then
         if [ "${TS_TUNE_MEMORY}" = "18446744073709551615" ]; then
             # Bash seems to error out for numbers greater than signed 64-bit,
             # so if the value of limit_in_bytes is the 64-bit UNSIGNED max value