improve assemble; update base and cloud-filesystem versions; disable quorum stuff by default

Author: williamstein

g-cloudfs

@@ -1 +1,5 @@
 make cloud-filesystem && make push-cloud-filesystem
+
+if [ `uname -m` = 'x86_64' ]; then
+   make assemble-cloud-filesystem
+fi

images.json

@@ -55,7 +55,10 @@
     "icon": "disk-round",
     "url": "https://github.com/sagemathinc/cocalc-compute-docker/tree/main/src/cloud-filesystem",
     "source": "https://github.com/sagemathinc/cocalc-compute-docker/tree/main/src/cloud-filesystem",
-    "versions": [{ "tag": "1.16.13", "tested": false }],
+    "versions": [
+      { "tag": "1.16.13", "tested": false },
+      { "tag": "1.16.14", "tested": false }
+    ],
     "description": "Highly scalable distributed POSIX Cloud File System built using JuiceFS, Google Cloud Storage and KeyDB."
   },
   "vpn": {
@@ -86,7 +89,8 @@
       { "tag": "1.1", "tested": true },
       { "tag": "1.2", "tested": true },
       { "tag": "1.3", "tested": true },
-      { "tag": "1.5.1", "tested": true }
+      { "tag": "1.5.1", "tested": true },
+      { "tag": "1.5.2", "tested": true }
     ],
     "description": "Base container that many others derive from."
   },

src/cloud-filesystem/scripts/cloud_filesystem.py

@@ -97,21 +97,32 @@
 # to close to this parameter.
 MAX_DATA_LOSS_S = 60
 
-# do not make this TOO small, e.g., it better be bigger than the typical unmount
-# and shutdown time
+# do not make this TOO small, e.g., it better be bigger than the
+# typical unmount and shutdown time
 MAX_REPLICA_LAG_S = max(10, MAX_DATA_LOSS_S - 10)
 
-MAX_LOG_SIZE_MB = 100
+# We need more testing with this first.
+# Right now it seems to cause juicefs to crash when one node is running
+# and we start another and it takes a while to start.  That's just
+# really annoying, obviously.  So for now we turn off this requirement,
+# and will implement something at a different level to avoid split brain.
+# In most cases we only have a single client at a time for cocalc right now.
+QUORUM_ENABLED_BY_DEFAULT = False
+
+# The logfiles can get really big, especially from keydb.
+# We trim that periodically to be at most this large.
+MAX_LOG_SIZE_MB = 50
 
 # when waiting for new state, we wait at most this
 # long to make sure we regularly have a chance to
 # do misc maintenance, e.g., mounting unmounted filesystems.
-MAX_WAIT_TIME_S = 45
+MAX_WAIT_TIME_S = 30
 
 # TODO: This is NOT at all tested or working yet.  Just something to work
 # on someday.
 ENABLE_FLASH = False
 
+
 ###
 # Utilities
 ###
@@ -1112,7 +1123,7 @@ def get_min_replicas_to_write(port):
 
 
 def get_quorum(filesystem, network):
-    enabled = filesystem.get('quorum', True)
+    enabled = filesystem.get('quorum', QUORUM_ENABLED_BY_DEFAULT)
     if not enabled:
         # quorum requirement is disabled
         return 1

src/scripts/assemble.sh

@@ -18,7 +18,7 @@ echo TAG=$TAG
 attempt=0
 until docker manifest create   $IMAGE:$TAG $IMAGE-x86_64:$ARCH_TAG $IMAGE-arm64:$ARCH_TAG --amend; do
   attempt=$((attempt+1))
-  if [[ $attempt -ge 30 ]]; then
+  if [[ $attempt -ge 50 ]]; then
     echo "Command failed after 30 attempts, exiting..."
     exit 1
   fi