daverigby

___ Begin jemalloc statistics ___
Version: "5.2.1-0-gea6b3e973b477b8061e0076bb257dbd7f3faa756"
Build-time option settings
  config.cache_oblivious: false
  config.debug: false
  config.fill: true
  config.lazy_lock: false
  config.malloc_conf: ""
  config.opt_safety_checks: false
  config.prof: true

daverigby

1. Download from rpmfind.net:

wget https://www.rpmfind.net/linux/epel/9/Everything/x86_64/Packages/b/byobu-5.133-8.el9.noarch.rpm

2. Install:

sudo yum install ./byobu-5.133-8.el9.noarch.rpm 

daverigby

Based on information from: https://github.com/jemalloc/jemalloc/wiki/Use-Case%3A-Heap-Profiling

Build jemalloc

RedHat & related flavours don't enable profiling in their binaries, so we need to build from source:

wget https://github.com/jemalloc/jemalloc/releases/download/5.3.0/jemalloc-5.3.0.tar.bz2
tar xjf jemalloc-5.3.0.tar.bz2
cd jemalloc-5.3.0
./configure --enable-stats --enable-prof
make -j16

daverigby

#!/usr/bin/env python3

"""Reads from stdin / arguments a sequence of lines which contain an
ISO 8601 timestamp, and outputs to stdout each line with the timestamp
with the delta from the previous line.

In other words, converts a file with absolute timstamps into a file
with relative timestamps since the previous line.

The first line cannot have a delta calculated, so it is reported as

daverigby

[Wed Mar 15 13:43:09 2023] AuxIoPool1: page allocation failure: order:0, mode:0x1000000(GFP_NOWAIT), nodemask=(null)
[Wed Mar 15 13:43:09 2023] AuxIoPool1 cpuset=/ mems_allowed=0
[Wed Mar 15 13:43:09 2023] CPU: 6 PID: 2760 Comm: AuxIoPool1 Not tainted 4.14.305-227.531.amzn2.aarch64 #1
[Wed Mar 15 13:43:09 2023] Hardware name: Amazon EC2 c6gd.2xlarge/, BIOS 1.0 11/1/2018
[Wed Mar 15 13:43:09 2023] Call trace:
[Wed Mar 15 13:43:09 2023]  dump_backtrace+0x0/0x190
[Wed Mar 15 13:43:09 2023]  show_stack+0x24/0x30
[Wed Mar 15 13:43:09 2023]  dump_stack+0x98/0xc8
[Wed Mar 15 13:43:09 2023]  warn_alloc+0x11c/0x1b4

daverigby

• Expressions on different stats - doesn't work due to extra name label which we need to discard:

sum without (name) (kv_ep_db_data_size_bytes{bucket="$bucket"}) / sum without (name) (kv_ep_db_file_size_bytes{bucket="$bucket"})

Heatmaps

Note: Set chart type to 'Heatmap' (top right), Series format to 'Heatmap'

• Heatmap of KV opcode latencies:

sum by(le) (rate(kv_cmd_duration_seconds_bucket{opcode="ADD", bucket="$bucket"}[5m]))

daverigby

--- /Users/dave/Documents/MBs/MB-53924-v8_dyld/v8-8.3-cb5/libv8.dylib.load-commands	2022-09-30 19:52:49.000000000 +0100
+++ /Users/dave/Documents/MBs/MB-53924-v8_dyld/v8-10.7.21/libv8.dylib.load-commands	2022-09-30 19:53:03.000000000 +0100
@@ -1,23 +1,23 @@
-/Users/dave/Documents/MBs/MB-53924-v8_dyld/v8-8.3-cb5/libv8.dylib:
+/Users/dave/Documents/MBs/MB-53924-v8_dyld/v8-10.7.21/libv8.dylib:
 Load command 0
       cmd LC_SEGMENT_64
-  cmdsize 632
+  cmdsize 712

daverigby

Results

TL;DR: Apple M1 Pro is ~3x faster to build compared to i9-8950HK (Coffee Lake).

MacBook Pro 2018

MacBook Pro (15-inch, 2018)
Processor: 2.9 GHz 6-Core Intel Core i9-8950HK CPU Memory: 16 GB 2400 MHz DDR4

155.7s

daverigby

• Change num_writer_threads from command-line

curl --user Administrator:asdasd -X POST --data "num_writer_threads=disk_io_optimized" http://localhost:9000/pools/default/settings/memcached/global

(Can also use num_reader_threads, and specify an exact number)

• Change number of shards from command-line:

curl -i -u Administrator:asdasd --data 'ns_bucket:update_bucket_props("default", [{extra_config_string, "max_num_shards=64"}]).' http://localhost:9000/diag/eval

daverigby

<?xml version="1.0" encoding="UTF-8"?>
<manifest>
  <remote name="gerrit" fetch="ssh://review.couchbase.org:29418/" />
  <remove-project name="build" />
  <project path="cbbuild" name="build" remote="gerrit" revision="refs/changes/28/155828/1" groups="notdefault,build" />
</manifest>