Use a lossy timeseries to predict when a filesystem will be full (PoC)

predict_disk_full

@@ -0,0 +1,64 @@
+#!/usr/bin/python3
+
+# PoC
+
+# add a new entry and remove a random old entry with a probability
+# depending on the target size
+
+import argparse
+import datetime
+import json
+import random
+import sys
+import time
+import os
+
+ap = argparse.ArgumentParser()
+
+ap.add_argument("--targetsize", type=int, default=1000)
+ap.add_argument("filesystem")
+
+args = ap.parse_args()
+
+st = os.statvfs(args.filesystem)
+
+blocks_reserved = st.f_bfree - st.f_bavail
+blocks_usable = st.f_blocks - blocks_reserved
+blocks_used = st.f_blocks -  st.f_bfree
+
+# Just for readability
+bytes_usable = blocks_usable * st.f_bsize
+bytes_used = blocks_used * st.f_bsize
+
+
+filename = args.filesystem.replace("/", "_") + ".bytes.json"
+
+ts = []
+try:
+    with open(filename) as f:
+        ts = json.load(f)
+except FileNotFoundError as e:
+    pass # Ok, will create
+
+now = time.time()
+ts.append((now, bytes_used, bytes_usable))
+if random.random() > 1 - len(ts) / args.targetsize:
+    i = random.randrange(0, len(ts))
+    ts.pop(i)
+
+with open(filename, "w") as f:
+    json.dump(ts, f)
+
+for d in reversed(ts):
+    bytes_predicted = bytes_used * bytes_used / d[1]
+    print("Predicted use", bytes_predicted,
+          "in", now - d[0], "seconds",
+          "at", datetime.datetime.fromtimestamp(now + now - d[0]))
+    if bytes_predicted > bytes_usable:
+        print("Predicted disk full",
+              "in", now - d[0], "seconds",
+              "at", datetime.datetime.fromtimestamp(now + now - d[0]))
+        sys.exit(1)
+else:
+    print("safe for", now - ts[0][0], "seconds",
+          "at", datetime.datetime.fromtimestamp(now + now - ts[0][0]))

predict_disk_full_home_bytes

@@ -0,0 +1,4 @@
+#!/bin/sh
+set -e
+cd $(dirname $0)
+./predict_disk_full /home

predict_disk_full_var_bytes

@@ -0,0 +1,4 @@
+#!/bin/sh
+set -e
+cd $(dirname $0)
+./predict_disk_full /var