Mailing-List: contact commits-help@airflow.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@airflow.incubator.apache.org
Date: Mon, 23 May 2016 09:23:12 +0000 (UTC)
From: "Bence Nagy (JIRA)" <jira@apache.org>
To: commits@airflow.incubator.apache.org
Message-ID: <JIRA.12972001.1463995339000.264793.1463995392933@Atlassian.JIRA>
In-Reply-To: <JIRA.12972001.1463995339000@Atlassian.JIRA>
References: <JIRA.12972001.1463995339000@Atlassian.JIRA> <JIRA.12972001.1463995339285@arcas>
Subject: [jira] [Created] (AIRFLOW-163) Running multiple LocalExecutor
 schedulers makes system load skyrocket
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
archived-at: Mon, 23 May 2016 09:23:17 -0000

Bence Nagy created AIRFLOW-163:
----------------------------------

             Summary: Running multiple LocalExecutor schedulers makes syste=
m load skyrocket
                 Key: AIRFLOW-163
                 URL: https://issues.apache.org/jira/browse/AIRFLOW-163
             Project: Apache Airflow
          Issue Type: Bug
    Affects Versions: Airflow 1.7.1
         Environment: EC2 t2.medium instance,=20
Docker `version 1.11.1, build 5604cbe`,=20
Host is `Linux ip-172-31-44-140 3.13.0-85-generic #129-Ubuntu SMP Thu Mar 1=
7 20:50:15 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux`,=20
Docker containers are built upon the `python:3.5` image,=20
LocalExecutor is used with two scheduler containers running
            Reporter: Bence Nagy
            Priority: Minor


I've been told on Gitter that this is expected currently, but thought I'd c=
reate an issue for it anyway.

See this screenshot of a task duration chart =E2=80=94 I launched a second =
scheduler for the 8:50 execution. The orange line represents a PostgresOper=
ator task (i.e. processing happens independent of airflow), while the other=
 lines represent data copying tasks that go through a temp file on the airf=
low host https://i.imgur.com/2tDKgKj.png

I'm seeing a system load of around 4.0-5.0 when processing tasks when one s=
cheduler is running, and 20.0-30.0 with two.

Running {{airflow scheduler --num_runs 3}} under yappi got me these results=
 when ordered by total time: http://pastebin.com/8TiEG4P3. I still have the=
 raw profiling data, let me know if another data extract would be useful.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)