From 7608b6f0040ada497f15b70b8a9e8b8aa9e1019d Mon Sep 17 00:00:00 2001
From: Reid 'arrdem' McKenzie <me@arrdem.com>
Date: Thu, 19 Aug 2021 23:45:15 -0600
Subject: [PATCH] Document benchmark results

---
 projects/jobq/README.md                   | 56 +++++++++++++++++++++++
 projects/jobq/benchmark.py                | 13 ++++++
 projects/jobq/src/python/jobq/__init__.py |  2 +-
 3 files changed, 70 insertions(+), 1 deletion(-)

diff --git a/projects/jobq/README.md b/projects/jobq/README.md
index d9b5a96..914faf5 100644
--- a/projects/jobq/README.md
+++ b/projects/jobq/README.md
@@ -1,3 +1,59 @@
 # Jobq
 
 Abusing sqlite3 as a job queue.
+
+## Benchmarks
+
+Benchmarks are extremely steady.
+Flushing a sqlite file to disk seems to be the limiting factor of I/O, and pipelining multiple message writes is undoubtably the way to go.
+However the purpose of the API is to use the sqlite file as the shared checkpoint between potentially many processes, so 'large' transactions are an antipattern.
+
+Tests suggest that this library is rock steady at 100 writes per sec. and 100 polls per sec. and completely bounded by sqlite controlled I/O as evidenced by using `":memory:"` which doesn't have to `fsync()`.
+
+``` shell
+$ bazel run :benchmark
+
+...
+
+Target //projects/jobq:benchmark up-to-date:
+  bazel-bin/projects/jobq/benchmark
+
+...
+
+Ran 'insert' 10000 times, total time 101.810816516 (s)
+  mean: 0.010148992843 (s)
+  median: 0.009474293 (s)
+  stddev: 0.006727934042954838 (s)
+  test overhead: 3.20888086e-05 (s)
+
+Ran 'poll' 10000 times, total time 100.482262487 (s)
+  mean: 0.0100152467857 (s)
+  median: 0.0095528585 (s)
+  stddev: 0.00821730176268304 (s)
+  test overhead: 3.2979463000000004e-05 (s)
+
+Ran 'append_event' 10000 times, total time 105.015296419 (s)
+  mean: 0.0104681294652 (s)
+  median: 0.009592544 (s)
+  stddev: 0.007321370576225584 (s)
+  test overhead: 3.34001767e-05 (s)
+
+Testing with :memory:
+Ran 'insert' 10000 times, total time 0.37031511 (s)
+  mean: 3.3595880100000005e-05 (s)
+  median: 2.96015e-05 (s)
+  stddev: 1.045088890675899e-05 (s)
+  test overhead: 3.4356309e-06 (s)
+
+Ran 'poll' 10000 times, total time 1.17148314 (s)
+  mean: 0.0001128911222 (s)
+  median: 9.7398e-05 (s)
+  stddev: 3.213524197973896e-05 (s)
+  test overhead: 4.2571917999999996e-06 (s)
+
+Ran 'append_event' 10000 times, total time 0.415490332 (s)
+  mean: 3.78861989e-05 (s)
+  median: 3.3019e-05 (s)
+  stddev: 1.1752889674795285e-05 (s)
+  test overhead: 3.6628343e-06 (s)
+```
diff --git a/projects/jobq/benchmark.py b/projects/jobq/benchmark.py
index 582ae18..9d4b854 100644
--- a/projects/jobq/benchmark.py
+++ b/projects/jobq/benchmark.py
@@ -10,6 +10,7 @@ from random import randint, choice
 import string
 from statistics import mean, median, stdev
 import tempfile
+import logging
 
 from jobq import JobQueue
 
@@ -75,7 +76,17 @@ def test_poll(q, reps):
     bench(poll, reps)
 
 
+def test_append(q, reps):
+    def append_event():
+        q.append_event(randint(1, reps), {"foo": "bar"})
+
+    bench(append_event, reps)
+
+
 if __name__ == "__main__":
+    # No logs
+    logging.getLogger().setLevel(logging.WARN)
+
     # Test params
     reps = 10000
     path = "/tmp/jobq-bench.sqlite3"
@@ -89,8 +100,10 @@ if __name__ == "__main__":
     q = JobQueue(path)
     test_insert(q, reps)
     test_poll(q, reps)
+    test_append(q, reps)
 
     print(f"Testing with :memory:")
     q = JobQueue(":memory:")
     test_insert(q, reps)
     test_poll(q, reps)
+    test_append(q, reps)
diff --git a/projects/jobq/src/python/jobq/__init__.py b/projects/jobq/src/python/jobq/__init__.py
index e0ac59a..bd6a963 100644
--- a/projects/jobq/src/python/jobq/__init__.py
+++ b/projects/jobq/src/python/jobq/__init__.py
@@ -116,7 +116,7 @@ WHERE
 ;
 """
 
-_POLL_SQL = """\
+_POLL_SQL = f"""\
 UPDATE `job`
 SET
     `events` = json_insert(events, '$[#]', json_array('job_state_advanced', json_object('old', json(state), 'new', json(:state), 'timestamp', CURRENT_TIMESTAMP)))