ReqAcct is all zero for HTTP/2 #2238

Closed
xcir opened this Issue Mar 1, 2017 · 2 comments

Comments

Projects
None yet
5 participants
@xcir
Contributor

xcir commented Mar 1, 2017

Expected Behavior

varnishncsa %I and %O option output correctly in HTTP/2 request.

Current Behavior

Request by HTTP/1.1.

# varnishncsa -F "%r I=%I O=%O"
GET http://***/munin.png HTTP/1.1 I=460 O=5160
GET http://***/munin.png HTTP/1.1 I=460 O=5156
GET http://***/munin.png HTTP/1.1 I=460 O=5160

Request by HTTP/2.0.

# varnishncsa -F "%r I=%I O=%O"
GET http://***/munin.png HTTP/2.0 I=0 O=0
GET http://***/munin.png HTTP/2.0 I=0 O=0
GET http://***/munin.png HTTP/2.0 I=0 O=0

Request by HTTP/2.0. (varnishlog)

# varnishlog -graw -ireqacct
        46 ReqAcct        c 0 0 0 0 0 0
     65551 ReqAcct        c 0 0 0 0 0 0
     65550 ReqAcct        c 0 0 0 0 0 0
        49 ReqAcct        c 0 0 0 0 0 0

Possible Solution

Set the value to ReqAcct.

Steps to Reproduce (for bugs)

See the Current Behavior above.

Context

There is no way to know total bytes.

Your Environment

  • Version used: hitch1.4.4 + varnish-trunk revision bffb15e
  • Operating System and version: Ubuntu 16.04.2 LTS (4.4.0-64-generic)

@fgsch fgsch changed the title from [trunk] varnishncsa %I and %O value is all zero for HTTP/2 to ReqAcct is all zero for HTTP/2 Mar 1, 2017

@bsdphk

This comment has been minimized.

Show comment
Hide comment
@bsdphk

bsdphk Mar 14, 2017

Contributor

This is on purpose for now, because we need to decide how to count things in H2 relative to H1 to give meaningful statistics.

The "protocol overhead" of H1 is nearly non-existent, but in H2 it can be a significant fraction of the traffic and we must decide how we "charge" that in VSL and VSC.

Contributor

bsdphk commented Mar 14, 2017

This is on purpose for now, because we need to decide how to count things in H2 relative to H1 to give meaningful statistics.

The "protocol overhead" of H1 is nearly non-existent, but in H2 it can be a significant fraction of the traffic and we must decide how we "charge" that in VSL and VSC.

@bsdphk

This comment has been minimized.

Show comment
Hide comment
@bsdphk

bsdphk Apr 6, 2017

Contributor

This shold be fixed by 8abfe28

Contributor

bsdphk commented Apr 6, 2017

This shold be fixed by 8abfe28

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment