summaryrefslogtreecommitdiff
path: root/script
diff options
context:
space:
mode:
authorDouglas Bagnall <douglas.bagnall@catalyst.net.nz>2019-03-28 16:07:48 +1300
committerAndrew Bartlett <abartlet@samba.org>2019-05-01 05:32:25 +0000
commit60620273dba1d7f7ff25710c5dd8fd6d32f2d149 (patch)
treefac078624688c017ddfda35645803e33b68f2c4a /script
parenta047e71bc7daea3855fc786c0e1c9d09878bba5f (diff)
downloadsamba-60620273dba1d7f7ff25710c5dd8fd6d32f2d149.tar.gz
dsdb/modules: a module to count attribute searches and results
The dsdb module stack can turn a simple search request into a complicated tree of sub-queries that include attributes not originally asked for and excluding those that were. The corresponding replies might contain unrequested attributes or (for good reasons, according to some module) hide requested ones. The entire stack is there to meddle and that is what is does. Except *this* module. It just counts. To understand dsdb performance it helps to have some idea what requests and replies are flying too and fro. This module, when inserted anywhere in the stack, counts the requests and replies passing through and the attributes they contain. This data is stored in on-disk tdbs in the private/debug directory. The module is not loaded by default. To load it you need to patch the source4/dsdb/samdb/ldb_modules/samba_dsdb.c and put "count_attrs" somewhere in the module lists in the samba_dsdb_init() function. For example, to examine the traffic between repl_meta_data and group_audit_log, you would do something like this around line 316: "subtree_delete", "repl_meta_data", + "count_attrs", "group_audit_log", "encrypted_secrets", and recompile. Samba will then write to a number of tdb files in the debug directory as requests and replies pass through. A simple script is included to read these files. Doing this: ./script/attr_count_read st/ad_dc/private/debug/debug/attr_counts_not_found.tdb will print a table showing how often various attritbutes were requested but not found (from the point of view of the module). A more sophisticated version of the script is coming in the next commit, but this one is included first because in its simplicity it documents the storage format reasonably well. The tdb keys are attribute names, and the values are uint32_t in machine native order. When the module is included in the stack there will be a very small decrease in performance. Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz> Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Diffstat (limited to 'script')
-rwxr-xr-xscript/attr_count_read27
1 files changed, 27 insertions, 0 deletions
diff --git a/script/attr_count_read b/script/attr_count_read
new file mode 100755
index 00000000000..69c7c63e729
--- /dev/null
+++ b/script/attr_count_read
@@ -0,0 +1,27 @@
+#!/usr/bin/env python3
+
+import sys
+import struct
+sys.path.insert(0, "bin/python")
+import tdb
+
+
+def main(filename):
+ db = tdb.Tdb(filename)
+ pairs = []
+ longest = 0
+ for k in db:
+ v = struct.unpack("I", db[k])[0]
+ pairs.append((v, k.decode('utf-8')))
+ longest = max(len(k), longest)
+
+ pairs.sort()
+ for v, k in pairs:
+ print("%*s: %7d" % (longest, k, v))
+
+
+if len(sys.argv) < 2:
+ print("Usage: attr_count_read <tdb-file>")
+ sys.exit(1)
+
+main(sys.argv[1])