Working with Handlers » History » Version 18

« Previous - Version 18/97 (diff) - Next » - Current version
Rich Karstens, 04/30/2013 08:52 AM

Working with Handler Programs

Requirements of a Handler Program

There are few requirements on a handler program. Essentially the handler must return data (or a return code) within the time allotted by the WSS (configured via handlerTimeout parameter), should return data via stdout on success and should return the specific exit codes (described below) on error. An optional error string may be returned via stderr.

Invocation, Return Codes and Data

The WSS invokes the handler program via a two stage process. In the Stage 1, the handler program is invoked with all the its attendant arguments. The WSS then waits for a response. This response can be either

  1. A *NIX return code (usually indicating an error)
  2. Data on stdout (interpreted as 'SUCCESS')

or, if the handler fails to return a return code or data via stdout, a timeout occurs and the handler program is aborted and the web service receives the appropriate error response.

If data appears, Stage 2 commences. At this point, the WSS returns an 'OK' HTTP response to the client and begins streaming whatever data the handler returns directly to the client via HTTP.

Command Line Arguments

Command line arguments corresponding to query parameter name / value pairs will be added to the invoking command line of the handler program. Only those parameters configured in param.cfg will be added. All extraneous query parameters will cause an exception to be thrown by the WSS. Each query parameter pair i.e. &quality=B will be translated into --quality B on the command line.

Double hyphen '--' standard always used
Query parameter spelling is not translated. I.e. the name of the parameter on the URL translated directly into the argument to the handler, e.g. &network=IU on the URL becomes --network IU on the command line.

Only parameters configured via param.cfg are passed as parameters to the handler [3 exceptions below]

  1. Except --username USER which is not a query parameter and is added by the WSS for authenticated access and is based on the result of the container's authentication procedures.
  2. Except --STDIN which is not a query parameter. It is added as a parameter if HTTP POST is used (rather than HTTP GET). This indicates to the handler that it should
    process stdin.

Also, the query parameter &nodata=[205|404] is not added to the command line and affects only the WSS response itself.

Errors and Return Codes

Handler return codes and their translation into HTTP Status codes should conform to the following specification

Exit Status HTTP STATUS Description
0 200 Successfully processed request, data returned via stdout
1 500 General error. An error description may be provided on stderr
2 204 No data. Request was successful but results in no data
3 400 Invalid or unsupported argument/parameter
4 413 Too much data requested

Note that Entry '4' above, HTTP Status 204, can be translated to a 404 based on both an application parameter configuration or overridden by the client with a specific query parameter nodata=....

Timing out.

Timeouts can occur at any point after the handler program is invoked. The WSS will terminate the handler program if it fails to return a *NIX return code (or data) within the timeout period during Stage 1 or if it fails to produce data during Stage 2 for a period of time in excess of the timeout period. If a timeout occurs, the handler program is terminated.

If the timeout occurs during Stage 1, an appropriate error response is sent to the client. If it occurs during Stage 2, the transaction is terminated and the connection closed.

Error Text

If the handler program writes text via stderr during Stage 1 and an error code is received by the WSS, the WSS will include the error text in the error response sent to the client.

Output Formats

The WSS should be configured for the same output format (MIME type) that the handler program produces. The WSS uses the output type to a) Tell the HTTP client in what format the data is [Content-Type] and b) in the suggested file name [Content-Disposition]. A conflict between the configuration and the handler's output will not cause any harm, but will confuse clients.

The output format can be overridden via the URL using the &output=FORMAT query parameter. As in the configuration in the service.cfg file, the allowable values are limited:

output value MIME type
xml application/xml
mseed application/vnd.fdsn.mseed
text text/plain
texttree text/plain
json application/json

These are the output types recognized by the WSS and does not imply what any given service or handler must support.

Environment Variables Set by the WSS

The WSS sets certain pre-defined environment variables for use (if needed) by a handler program. The general intention of these variables are to allow the handler program to know about the HTTP request and the version of the WSS used with any eye towards logging. I.e. if a handler program wished to perform its own 'per request' logging, this information would be vital. Below is a table with the environment variables.

Varible name Value
REQUESTURL The URL of the incoming request
USERAGENT User agent string supplied in the HTTP header for this request
IPADDRESS IP Address of the request's client
APPNAME Application name supplied via the appName parameter from the service configuration
VERSION Application version supplied via the version parameter from the service configuration

Handling Network Interruptions

It is somewhate common for the network connection from the client to the handler to be interrupted. This can occur due to the network connection being dropped, some problem in any intermediate network stage, or the client closing the connection while a transfer is ongoing. One common example is if the client is a browser, the user requests a SEED file via dataselect and then dismissed the File Save dialog via 'Cancel'.

A handler program should gracefully handle a network disconnection, clean up and exit. Often these disconnections are detected via an IO Exception when the stdin connection to which the handler had been writing goes away. Defensive programming is always preferred, but, in general, nothing untoward happens with regard to the WSS if the handler behaves badly. If the handler stops sending data for a long enough length of time, the WSS will terminate it. If the interruption is upstream from the WSS, i.e the client appears to disconnect, the current invocation of the WSS will attempt to terminate its handler invocation. This prevents zombie handler processes.