twisted.web server keeps caching headers if an exception is raised

	ivank reported
Trac ID	trac#4897
Type	defect
Created	2011-02-18 02:06:35Z

If a resource sets caching headers but then raises an exception, twisted.web.server serves an error page with the existing caching headers. This may cause the error page to be cached.

For example:

def render_GET(self, request):
    request.responseHeaders.setRawHeaders('cache-control', ['max-age=172800, private'])
    request.responseHeaders.setRawHeaders('expires', ['Sun, 20 Feb 2011 01:57:40 GMT'])
    if random.random() < 0.05:
        # Pretend this is a database failure or something
        1/0

In this case, the error page served by twisted.web.server.processingFailed might be cached.

If this problem can be solved by clearing or overriding cache-related headers, twisted.web could help users avoid a non-obvious pitfall. If the right thing to do is to keep the cache-related headers, this behavior should be documented so that users know to catch their own exceptions.

trac-id__4897 4897
type__defect defect
reporter__ivank ivank
priority__normal normal
milestone__ 
branch__ 
branch_author__ 
status__new new
resolution__None None
component__web web
keywords__ 
time__1297994795000000 1297994795000000
changetime__1298066267000000 1298066267000000
version__None None
owner__ 
cc__jknight cc__ivank

Start by inspecting twisted.web.server.processingFailed and the error path described in the report. Reproduce the resource example, then establish whether the intended outcome is to clear or override cache-related headers or to document the existing behavior; done means the chosen behavior is verified for an exception response.