-
Notifications
You must be signed in to change notification settings - Fork 655
WebJobs
Ability to run a script (called job) in 2 different ways:
- Trigger - run the job every time an API is called or a schedule is reached.
- Continuous - make sure the job is always running (job should have a while loop but when it goes down we bring it back up).
Jobs are deployed by copying them to d:\home\site\wwwroot
as illustrated below. These Jobs will run under the worker process of the SCM\Kudu Web Site and not the main site.
d:\
|-- home
|-- site
|-- wwwroot
|-- App_Data
|-- jobs
|-- triggered
| |-- job1
| | |-- run.bat
| | |-- other files
| |-- job2
| |-- run.bat
| |-- other files
|-- continuous
|-- job3
| |-- run.bat
| |-- other files
|-- job4
|-- run.bat
|-- other files
To deploy a webjob, create a zip file with content structure as it will be deflated to d:\home\site\wwwroot
as is - for instance, the top level folder when you open a zip file should be App_Data folder. Then, simply use zipdeploy to deploy the zip. This is commonly used to deploy webjobs via ARM template or EV2.
As an alternative, d:\home\site\jobs\...
can be used instead of d:\home\site\wwwroot\app_data\jobs\...
. This is useful when using Run-From-Package, which makes d:\home\site\wwwroot
become read-only.
Managing Azure WebJobs via the portal with Run from Package
Customers who are running with the default location of d:\home\site\wwwroot\app_data\jobs\...
and using Run-From-Package may see 404 responses when starting or stopping the WebJobs. The start\stop mechanism requires write access to add the disable.job for stopping a continuous WebJob and deleting it when starting it. This will not succeed when the location is read-only.
Note: see the Deployment section below for alternative ways of deploying WebJobs.
The following file types are accepted as runnable scripts that can be used as a job:
- .fsx (using the F# fsi.exe interactive compiler)
- .cmd, .bat, .exe (using windows cmd)
- .ps1 (using powershell)
- .sh (using bash)
- .php (using php)
- .py (using python)
- .js (using node)
- project.json (using Dnx - details here)
We use the following logic to decide which file is the script to run within the job's directory:
- Per file type we look first for a file named:
run.{file type extension}
(for examplerun.cmd
orrun.exe
). - If it doesn't exists for all file types, we'll then look for the first file with a supported file type extension.
- The order of file types extension used is:
.cmd
,.bat
,.exe
,.ps1
,.sh
,.php
,.py
,.js
. - The recommended script file to have in your job directory is:
run.cmd
. - Note: We'll only look for a script under the root directory of that job (not under sub directories of it).
- Note: make sure .bat/.cmd files don't include the UTF-8 BOM (inserted by some editors such as Visual Studio by default), which can break things!
- WEBJOBS_RESTART_TIME - Timeout in seconds between when a continuous job's process goes down (for any reason) and the time we re-launch it again (Only for continuous jobs).
- WEBJOBS_IDLE_TIMEOUT - Time in seconds after which we'll abort a running triggered job's process if it's in idle, has no cpu time or output (Only for triggered jobs).
- WEBJOBS_HISTORY_SIZE - Maximum number of runs kept in the history directory for a triggered job. The default is 50.
- WEBJOBS_STOPPED - Set this setting to 1 to disable running any job (will also stop all currently running jobs).
-
WEBJOBS_DISABLE_SCHEDULE - Set this to 1 to turn off all scheduled triggering. Unlike with
WEBJOBS_STOPPED
, WebJobs can still be manually invoked. -
WEBJOBS_ROOT_PATH - Location of webjob files, you can specify an absolute path, or otherwise the value will be combined with the default root path:
D:/home/site/wwwroot/
+WEBJOBS_ROOT_PATH(relative)
-
WEBJOBS_LOG_TRIGGERED_JOBS_TO_APP_LOGS - If set to
true
, output from triggered WebJobs get sent to the Application Logs pipeline (which supports file system, blobs and tables)
When a job is invoked, several settings are added to its environment that the job process can use:
- WEBJOBS_PATH - The root path of currently running job (will be under some temporary directory).
- WEBJOBS_NAME - The current job name.
- WEBJOBS_TYPE - The current job type (triggered/continuous).
- WEBJOBS_DATA_PATH - The current job meta data path, this includes the job's logs/history and any artifact of the job can go there.
- WEBJOBS_RUN_ID - The current run id of the job (used for triggered jobs).
For continuous WebJobs - Console.Out and Console.Error are routed to the "application logs", they will show up as file, blob or table storage depends on your configuration of the application logs (similar to your Website).
Also the first 100 lines in each invocation will also go to the WebJob log file (to ease debugging pain when the WebJob fails at startup) accessible using the Azure portal (but also saved on your site's file system at data/jobs/continuous/jobName).
For triggered WebJobs - Console.Out/Console.Error are routed to the WebJobs specific run log file also accessible using the Azure portal and stored at data/jobs/triggered/jobName/runId. To have them also get routed to "application logs", set the WEBJOBS_LOG_TRIGGERED_JOBS_TO_APP_LOGS
to true
.
Console.Out is treated (marked) as INFO and Console.Error as ERROR.
There are multiple ways to deploy WebJobs.
From the portal, you can directly upload a zip file that contain the WebJob's files.
You can deploy a WebJob by directly copying the files under the locations discussed above (i.e. wwwroot/app_data/jobs/...
). This can be done either by FTP, WebDeploy, Kudu Console or git.
If you have Visual Studio 2013 Update 3 (or later), it lets you do exactly this (more info here). Previously, this was done using the WebJobsVs VS add-in, but that is now obsolete.
High level:
- You have both an ASP.NET app and one or more WebJobs (Console apps) in your solution
- You ask the ASP.NET app to include the WebJobs apps with itself when deploying
This works both when deploying directly from Visual Studio (WebDeploy), or via git.
The powershell Azure RM cmdlets can be used to deploy a WebJob. See the example script here.
Graceful shutdown is the ability to notify the job process that it's going to shut down and wait a period of time for the process to go down by itself before killing it.
The graceful shutdown feature works slightly different for continuous and triggered jobs.
When a shutdown request is detected a file will be created in the path: %WEBJOBS_SHUTDOWN_FILE%
(environment variable which can be obtained by the job's process).
It is then the WebJobs process's responsibility to detect the presence of this file and initiate shutdown. Note that when using the WebJobs SDK, that part is done automatically, so you don't need to worry about this file.
Then there is a default period of 5 seconds waiting for the job process to shutdown before getting killed.
When a shutdown request is detected there is a 30 seconds default waiting period for the job process to stop.
You can change the grace period of a job by specifying it (in seconds) in the settings.job file where the name of the setting is stopping_wait_time like so:
{ "stopping_wait_time": 60 }
This sample will have the job grace period at 60 seconds instead of the default (but see hard shutdown discussion below).
Note that when using the WebJobs SDK, it automatically listens to %WEBJOBS_SHUTDOWN_FILE%
, and will terminate the process if no functions are running. WHen that happens, stopping_wait_time
is often irrelevant, as it only applies if the process refuses to shut down on its own despite the presence of %WEBJOBS_SHUTDOWN_FILE%
.
There are several scenarios that can trigger a WebJobs shutdown. In some cases, it is possible to extend to shut down period indefinitely (soft), while in others it is limited (see Hard shutdowns below). What determines whether a shutdown is 'hard' is whether the Web App itself needs to restart.
Hard shutdowns:
- The Web App is restarted and the shutdown time limit is 30s.
- The VM the Web App lives on needs to be upgraded, and the Web App needs to be moved to a different VM and the shutdown time limit is 3 mins.
- The App Service Plan's instance count was reduced and some VMs need to be removed and the shutdown time limit is 3 mins.
Soft shutdowns:
- The WebJob is stopped (only applies to continuous WebJobs)
- The WebJobs files are updated
There are 2 options for where the WebJob is running from (what is the working directory):
- The WebJob is copied to a temporary directory under
%TEMP%\jobs\{job type}\{job name}\{random name}
and will run from there This option prevents the original WebJob binaries from being locked which might cause issues redeploying the WebJob. For example updating an .exe file that is currently running. - The WebJob is run directly from the WebJob binaries directory. We call this option in place.
This option can have the locking issue and should only be used when there is no risk of locking the file.
For example if the WebJob only consists of a
.cmd
file.
By default the first option is used (in place = false
) other than for .js
node.js scripts (in place = true
).
You can explicitly configure this setting in your settings.job. e.g.
{ "is_in_place": true/false }
To schedule a triggered WebJob you need to add a schedule property to the settings.job file.
The value of the schedule is cron expression that has 6 fields to represent the schedule: {second} {minute} {hour} {day} {month} {day of the week}
.
Each field can have a specific value (1), a range (1-10), a set of values (1,2,3), all values (), an interval value (/2 == 0,2,4,6,...) or a mix of these (1,5-10).
Whenever the settings.job
file is updated with a new schedule, the WebJob's schedule will automatically update, it is best to deploy the settings.job
as part of the WebJob.
For the schedule to work it requires the website to be configured as Always On and is not an Azure Scheduler but an internal implementation of a scheduler.
The local time is, by default, the timezone of server machine. In Azure environment, by default, it is UTC. To adjust the timezone, you may specify the appSetting
WEBSITE_TIME_ZONE
with the value of the name of timezone; for instance,Pacific Standard Time
. The worker process as well as webjobs will use that as local time zone.
Here is an example schedule that will run once every minute:
{
"schedule": "0 * * * * *"
}
-
0 */5 * * * *
- run every 5 minutes (specifically on minute 0, 5, 10, 15, ...). -
0 0 0 * * 1-5
- run at midnight on Monday, Tuesday, Wednesday, Thursday and Friday.
Note samples are missing the seconds field, to make them work just add
0
as the first field to make them 6 fields total.
In Azure environment, the scheduled webjob will only run on one instance (or worker). This is achieved by using file lock mechanism while the webjob is running. If the webjob happens to run for a very short time, there could be a race condition (such as clock sku) to cause another instance to later start up on the same schedule (since it detects no lock). The workaround is to make sure the webjob run for a certain period of time (say at least 5 seconds assuming schedule interval is greater).
An optional settings.job
file can be deployed with a WebJob to tweak various settings. This file must be located at the root of the WebJobs's folder, along side your WebJob's script (e.g. in wwwroot\app_data\jobs\triggered\{job name}
or wwwroot\app_data\jobs\continuous\{job name}
).
settings.job
is a json file with top level settings, e.g.
{
"schedule": "0 * * * * *",
"is_in_place": true
}
The following settings are supported (click setting name for detailed information).
Settings usable on all WebJobs:
- is_in_place: allows the job to run in place without being first copied to a temp folder
- stopping_wait_time: allows control of the shutdown behavior
Settings usable only on Continuous WebJobs:
- is_singleton: only run the WebJobs on a single instance when scaled out
Settings usable only on Triggered WebJobs:
- schedule: run the WebJob on a CRON schedule