Update zfs_probe.py

zfs_probe.py

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
 
-# example: python3 zfs_probe.py --duration 10 --interval 1 --track-files
+# example: python3 zfs_probe.py --duration 120 --interval 5 --track-files --path-prefix /home
 
 import argparse
 import ast
@@ -223,6 +223,110 @@ def zfs_get_properties(pools):
     return rows
 
 
+
+
+def get_zfs_mountpoints():
+    """
+    Zwraca mountpointy filesystemów typu zfs widoczne w namespace procesu.
+    /proc/self/mountinfo opisuje mounty w bieżącym mount namespace.
+    """
+    mountinfo = "/proc/self/mountinfo"
+    mounts = []
+    text = safe_read_text(mountinfo)
+    for line in text.splitlines():
+        line = line.strip()
+        if not line or " - " not in line:
+            continue
+        left, right = line.split(" - ", 1)
+        lparts = left.split()
+        rparts = right.split()
+        if len(lparts) < 5 or len(rparts) < 1:
+            continue
+        fstype = rparts[0]
+        mountpoint = lparts[4]
+        if fstype == "zfs":
+            mounts.append(mountpoint)
+
+    # najdłuższe prefiksy najpierw, bez duplikatów
+    uniq = sorted(set(mounts), key=lambda x: (-len(x), x))
+    return uniq
+
+
+def path_is_under_mountpoints(path, mountpoints):
+    if not path or not path.startswith("/"):
+        return False
+    norm = os.path.normpath(path)
+    for mp in mountpoints:
+        if mp == "/":
+            return True
+        if norm == mp or norm.startswith(mp.rstrip("/") + "/"):
+            return True
+    return False
+
+
+def filter_bpf_opens_to_zfs(opens_map, mountpoints):
+    kept = {}
+    dropped_relative = 0
+    dropped_non_zfs = 0
+
+    for path, value in opens_map.items():
+        if not isinstance(path, str):
+            continue
+        if not path.startswith("/"):
+            dropped_relative += value
+            continue
+        if path_is_under_mountpoints(path, mountpoints):
+            kept[path] = kept.get(path, 0) + value
+        else:
+            dropped_non_zfs += value
+
+    return kept, dropped_relative, dropped_non_zfs
+
+
+def normalize_prefixes(prefix_arg):
+    prefixes = []
+    for item in (prefix_arg or "").split(","):
+        item = item.strip()
+        if not item:
+            continue
+        norm = os.path.normpath(item)
+        if not norm.startswith("/"):
+            continue
+        prefixes.append(norm)
+    return sorted(set(prefixes), key=lambda x: (-len(x), x))
+
+
+def path_matches_prefixes(path, prefixes):
+    if not path or not path.startswith("/"):
+        return False
+    norm = os.path.normpath(path)
+    for prefix in prefixes:
+        if prefix == "/":
+            return True
+        if norm == prefix or norm.startswith(prefix.rstrip("/") + "/"):
+            return True
+    return False
+
+
+def filter_bpf_opens_to_prefixes(opens_map, prefixes):
+    kept = {}
+    dropped_relative = 0
+    dropped_outside = 0
+
+    for path, value in opens_map.items():
+        if not isinstance(path, str):
+            continue
+        if not path.startswith("/"):
+            dropped_relative += value
+            continue
+        if path_matches_prefixes(path, prefixes):
+            kept[path] = kept.get(path, 0) + value
+        else:
+            dropped_outside += value
+
+    return kept, dropped_relative, dropped_outside
+
+
 def parse_arcstats():
     path = "/proc/spl/kstat/zfs/arcstats"
     if not os.path.exists(path):
@@ -257,11 +361,6 @@ def arc_delta(prev, curr):
 
 
 def parse_zpool_iostat_once(pool_list, interval):
-    """
-    Działa z różnymi wersjami zpool iostat.
-    Próbuje najpierw z -l, potem bez -l.
-    Nie zakłada obecności timestampa.
-    """
     candidate_cmds = [
         ["zpool", "iostat", "-H", "-p", "-y", "-l"] + pool_list + [str(interval), "1"],
         ["zpool", "iostat", "-H", "-p", "-y"] + pool_list + [str(interval), "1"],
@@ -362,30 +461,75 @@ def top_entries(entries, key, topn):
     return vals[:topn]
 
 
+def parse_bpftrace_version(text):
+    m = re.search(r'v?(\d+)\.(\d+)(?:\.(\d+))?', text)
+    if not m:
+        return None
+    major = int(m.group(1))
+    minor = int(m.group(2))
+    patch = int(m.group(3) or 0)
+    return (major, minor, patch)
+
+
+def choose_bpftrace_field_style():
+    """
+    bpftrace 0.19 zmienił składnię z args->field na args.field.
+    Dla 0.16 używamy starej składni, dla nowszych nowych wersji - nowej.
+    """
+    if not command_exists("bpftrace"):
+        return None, None, "bpftrace pominięty: brak polecenia bpftrace."
+
+    rc, out, err = run_cmd(["bpftrace", "--version"], timeout=10)
+    text = (out or err or "").strip()
+    version = parse_bpftrace_version(text)
+    if version is None:
+        return None, text, "bpftrace wykryty, ale nie udało się odczytać wersji z: {0}".format(text)
+
+    if version < (0, 19, 0):
+        return "arrow", text, None
+    return "dot", text, None
+
+
+def build_bpftrace_program(track_files=False, field_style="dot"):
+    if field_style == "arrow":
+        ret_expr = "args->ret"
+        bytes_expr = "args->bytes"
+        filename_expr = "str(args->filename)"
+    else:
+        ret_expr = "args.ret"
+        bytes_expr = "args.bytes"
+        filename_expr = "str(args.filename)"
+
+    lines = []
+    lines.append("tracepoint:syscalls:sys_exit_read /{0} > 0/ {{ @read_bytes_by_comm[comm] = sum({0}); @read_calls_by_comm[comm] = count(); }}".format(ret_expr))
+    lines.append("tracepoint:block:block_rq_issue {{ @block_bytes_by_comm[comm] = sum({0}); @block_ios_by_comm[comm] = count(); }}".format(bytes_expr))
+    if track_files:
+        lines.append("tracepoint:syscalls:sys_enter_openat {{ @opens[{0}] = count(); @opens_by_comm[comm, {0}] = count(); }}".format(filename_expr))
+    lines.append("END {")
+    lines.append('  printf("===READ_BYTES_BY_COMM===\\n"); print(@read_bytes_by_comm);')
+    lines.append('  printf("===READ_CALLS_BY_COMM===\\n"); print(@read_calls_by_comm);')
+    lines.append('  printf("===BLOCK_BYTES_BY_COMM===\\n"); print(@block_bytes_by_comm);')
+    lines.append('  printf("===BLOCK_IOS_BY_COMM===\\n"); print(@block_ios_by_comm);')
+    if track_files:
+        lines.append('  printf("===OPENS===\\n"); print(@opens);')
+        lines.append('  printf("===OPENS_BY_COMM===\\n"); print(@opens_by_comm);')
+    lines.append("}")
+    return "\n".join(lines) + "\n"
+
+
 def start_bpftrace(outdir, track_files=False):
     if os.geteuid() != 0:
         return None, "bpftrace pominięty: uruchom jako root."
-    if not command_exists("bpftrace"):
-        return None, "bpftrace pominięty: brak polecenia bpftrace."
 
-    program = []
-    program.append('tracepoint:syscalls:sys_exit_read /args.ret > 0/ { @read_bytes_by_comm[comm] = sum(args.ret); @read_calls_by_comm[comm] = count(); }')
-    program.append('tracepoint:block:block_rq_issue { @block_bytes_by_comm[comm] = sum(args.bytes); @block_ios_by_comm[comm] = count(); }')
-    if track_files:
-        program.append('tracepoint:syscalls:sys_enter_openat { @opens[str(args.filename)] = count(); @opens_by_comm[comm, str(args.filename)] = count(); }')
-    program.append('END {')
-    program.append('  printf("===READ_BYTES_BY_COMM===\\n"); print(@read_bytes_by_comm);')
-    program.append('  printf("===READ_CALLS_BY_COMM===\\n"); print(@read_calls_by_comm);')
-    program.append('  printf("===BLOCK_BYTES_BY_COMM===\\n"); print(@block_bytes_by_comm);')
-    program.append('  printf("===BLOCK_IOS_BY_COMM===\\n"); print(@block_ios_by_comm);')
-    if track_files:
-        program.append('  printf("===OPENS===\\n"); print(@opens);')
-        program.append('  printf("===OPENS_BY_COMM===\\n"); print(@opens_by_comm);')
-    program.append('}')
+    field_style, version_text, version_note = choose_bpftrace_field_style()
+    if version_note:
+        return None, version_note
+
+    program = build_bpftrace_program(track_files=track_files, field_style=field_style)
 
     bt_path = os.path.join(outdir, "trace.bt")
     with open(bt_path, "w") as f:
-        f.write("\n".join(program) + "\n")
+        f.write(program)
 
     out_path = os.path.join(outdir, "bpftrace.txt")
     out_f = open(out_path, "w")
@@ -395,7 +539,19 @@ def start_bpftrace(outdir, track_files=False):
         stderr=subprocess.STDOUT,
         universal_newlines=True,
     )
-    return {"proc": proc, "out_f": out_f, "out_path": out_path, "bt_path": bt_path}, None
+
+    note = "bpftrace: {0}, skladnia: {1}".format(
+        version_text if version_text else "unknown",
+        "args->field" if field_style == "arrow" else "args.field",
+    )
+    return {
+        "proc": proc,
+        "out_f": out_f,
+        "out_path": out_path,
+        "bt_path": bt_path,
+        "version_text": version_text,
+        "field_style": field_style,
+    }, note
 
 
 SECTION_RE = re.compile(r"^===([A-Z0-9_]+)===$")
@@ -461,7 +617,10 @@ def stop_bpftrace(handle):
         return
     proc = handle["proc"]
     if proc.poll() is not None:
+        try:
             handle["out_f"].close()
+        except Exception:
+            pass
         return
 
     try:
@@ -491,6 +650,8 @@ def main():
     ap.add_argument("--top", type=int, default=15, help="Ile pozycji pokazać w topkach")
     ap.add_argument("--outdir", default="", help="Katalog na logi i JSON")
     ap.add_argument("--track-files", action="store_true", help="Śledź top otwierane pliki przez bpftrace")
+    ap.add_argument("--zfs-files-only", action="store_true", help="W topce plików pokazuj tylko absolutne ścieżki leżące na mountpointach ZFS")
+    ap.add_argument("--path-prefix", default="", help="Pokaż tylko pliki spod tego prefiksu lub kilku prefiksów rozdzielonych przecinkami, np. /home/tank,/mnt/data")
     ap.add_argument("--no-bpf", action="store_true", help="Nie uruchamiaj bpftrace nawet gdy jest dostępny")
     args = ap.parse_args()
 
@@ -512,6 +673,12 @@ def main():
         sys.exit(1)
 
     oprint("Pule: {0}".format(", ".join(pools)))
+    zfs_mountpoints = get_zfs_mountpoints()
+    if zfs_mountpoints:
+        oprint("Mountpointy ZFS: {0}".format(", ".join(zfs_mountpoints)))
+    path_prefixes = normalize_prefixes(args.path_prefix)
+    if path_prefixes:
+        oprint("Filtr ścieżek: {0}".format(", ".join(path_prefixes)))
     meta = {
         "started_at": datetime.now().isoformat(),
         "duration": args.duration,
@@ -519,6 +686,8 @@ def main():
         "pools": pools,
         "hostname": os.uname().nodename,
         "uid": os.geteuid(),
+        "zfs_mountpoints": zfs_mountpoints,
+        "path_prefixes": path_prefixes,
     }
     with open(os.path.join(outdir, "meta.json"), "w") as f:
         json.dump(meta, f, indent=2)
@@ -529,6 +698,8 @@ def main():
         f.write(zpool_history_text(pools))
     with open(os.path.join(outdir, "zfs_properties.json"), "w") as f:
         json.dump(zfs_get_properties(pools), f, indent=2)
+    with open(os.path.join(outdir, "zfs_mountpoints.json"), "w") as f:
+        json.dump(zfs_mountpoints, f, indent=2)
 
     bpf_handle = None
     bpf_note = None
@@ -618,6 +789,9 @@ def main():
     oprint()
     oprint("Wyniki zapisane w: {0}".format(outdir))
     oprint("Pule: {0}".format(", ".join(pools)))
+    zfs_mountpoints = get_zfs_mountpoints()
+    if zfs_mountpoints:
+        oprint("Mountpointy ZFS: {0}".format(", ".join(zfs_mountpoints)))
     oprint("Czas zbierania: {0}s, interwał: {1}s, próbek: {2}".format(
         args.duration, args.interval, sum(len(v) for v in samples_by_pool.values())
     ))
@@ -701,10 +875,30 @@ def main():
 
         opens = bpf.get("OPENS", {})
         if opens:
+            report_opens = opens
+            title = "Top otwierane pliki z bpftrace"
+
+            if path_prefixes:
+                report_opens, dropped_relative, dropped_outside = filter_bpf_opens_to_prefixes(opens, path_prefixes)
+                title = "Top otwierane pliki z bpftrace dla wskazanych ścieżek"
+                oprint("Filtrowanie plików po prefiksie:")
+                oprint("  zostawione wpisy: {0}".format(sum(report_opens.values()) if report_opens else 0))
+                oprint("  odrzucone ścieżki względne: {0}".format(dropped_relative))
+                oprint("  odrzucone poza prefiksem: {0}".format(dropped_outside))
+                oprint()
+            elif args.zfs_files_only:
+                report_opens, dropped_relative, dropped_non_zfs = filter_bpf_opens_to_zfs(opens, zfs_mountpoints)
+                title = "Top otwierane pliki na ZFS z bpftrace"
+                oprint("Filtrowanie plików:")
+                oprint("  zostawione wpisy na ZFS: {0}".format(sum(report_opens.values()) if report_opens else 0))
+                oprint("  odrzucone ścieżki względne: {0}".format(dropped_relative))
+                oprint("  odrzucone ścieżki poza ZFS: {0}".format(dropped_non_zfs))
+                oprint()
+
             rows = []
-            for path, value in sorted(opens.items(), key=lambda kv: kv[1], reverse=True)[:args.top]:
+            for path, value in sorted(report_opens.items(), key=lambda kv: kv[1], reverse=True)[:args.top]:
                 rows.append({"path": path, "opens": value})
-            print_table("Top otwierane pliki z bpftrace", rows, [
+            print_table(title, rows, [
                 ("path", "path"),
                 ("opens", "opens"),
             ])
@@ -715,10 +909,12 @@ def main():
         "zpool_status_end.txt",
         "zpool_history.txt",
         "zfs_properties.json",
+        "zfs_mountpoints.json",
         "samples_zpool.json",
         "samples_arc.json",
         "proc_totals.json",
         "bpftrace.txt",
+        "trace.bt",
     ]:
         path = os.path.join(outdir, name)
         if os.path.exists(path):