It was counter-intuitive that, for pass objects, we logged the
parsed TTL when in fact the ttl visible from vcl was always 0.
In the builtin.vcl the Hit-For-Pass check was also called
for objects already classified as uncacheable.

Now, for passes, we

- do not parse Cache-Control / Expires
  - and, consequently, do not log TTL RFC
- do not call the hfp-code in builtin.vcl
  - and, consequently, do not log TTL VCL

Reflect this in b2.vtc

called

For some (out of workspace or stevedore allocationi failure) errors
VFP_Close would fail to be called when ending up not retrying the
fetch. This would then potentially leak VFP entry context data like
gzip buffers. Fix this by always calling VFP_Close() when changing
fetch state when VFP_Open() has been called.

Also VFP_Close() would for some retries be called twice, causing
double accounting log records print out. This happens when a late
error is caught and streaming is disabled, allowing a retry to be
attempted. Fix this by not unconditionally calling VFP_Close() in the
error state.

Also spring some VUT_Error() love.

Refs #2008

The child_poker lives in the manager process, and sends pings to the
child. With this patch we check that we actually get a PONG back, and
not some random data. If problems are detected, we kill the child.

Related to: #2010

Thx @fgs

Thx @fgs

We're only holding a read lock on the director, but we're updating
the nxt member concurrently. This should be acceptable as a performance
tradeoff - the only consequence is that round-robin is not strictly
going around - it may occasionally skip a backend or hand out the same
multiple times in a row.

the race is:

	thread	code

	A:	rr->nxt %= rr->vd->n_backend;
	// rr->nxt == rr->vd->n_backend - 1
	B:	rr->nxt++;
	// rr->nxt == rr->vd->n_backend
	A:	be = rr->vd->backend[nxt];
	// BOOM

should fix #2024

This is particularly useful when replaying vsl files (-r option).

The new -B option specifies a time factor limiting the speed at which
the log is being read. It can also be halved/doubled using the < and >
keys at runtime.

Actually implement the documented -p option

Make the update delay a double to allow update frequencies of less
than a second.

Add + and - keys to the curses interface to update the refresh
frequency while running.

Show update frequency in the top line.

... for the standard backend timestamps and size.

In usage/documentation, mark those options with <> which are not a
literal string, but rather a placeholder for some value.

Reported by Michel Schuster.

Previously you needed an empty argument (''), which didn't play very
well along sysv init scripts and was not documented.
This should address both problems.

Fixes #2007.

Tunable madvise for file storage

Adresses a new issue mentioned in #2004

Spotted by Foxik on #varnish @ irc.linpro.no.

Fixes #2011

Ref: #2008

We will need it if the body was consumed due to a call to
std.cache_req_body (VRB_Cache).

Analysis and test by daghf.

Fixes #1881.