- 23 Apr, 2019 11 commits
-
-
Nils Goroll authored
includes vtc Purposes: * Allow to optimize keepalive for trusted client connections (e.g. CDN, authenticated clients) * Prepare for an optimization to time out connections on inactive VCLs earlier (see #2764) Other client timeouts will be added when accepted.
-
Nils Goroll authored
Fixes #2949
-
Martin Blix Grydeland authored
This rushes the waitinglist when encountering a hit-for-pass during HSH_Lookup. Without this, the rushing may be delayed until the passed object completes delivery. Test cases c00098.vtc and c00099.vtc tests this early rushing behaviour for HFP and HFM respectively. Fixes: #2978
-
Martin Blix Grydeland authored
This makes waitinglist rushes happen also in HSH_Lookup when encountering cache hits. This helps to get the requests on the waitinglist restarted when doing streaming delivery. Fixes: #2977
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
their VCL from going cold. One half of #2471
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
which nobody else is allowed to even think about calling.
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
- 22 Apr, 2019 1 commit
-
-
Nils Goroll authored
-
- 21 Apr, 2019 2 commits
-
-
Federico G. Schwindt authored
ca-certificates is already installed so for now just ignore any ssl errors.
-
Federico G. Schwindt authored
Hopefully this will fix the error with coverity.
-
- 20 Apr, 2019 5 commits
-
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
-
Nils Goroll authored
the pointer indirection would make sense if we had any long lived struct waited instances so the timeout parameter could change. But we don't. Also the vtim_ type was wrong, it is a duration.
-
Nils Goroll authored
-
Federico G. Schwindt authored
Via lgtm.
-
- 18 Apr, 2019 11 commits
-
-
Nils Goroll authored
exotic compilers complain about incomplete type
-
Geoff Simmons authored
It had been iterating through the vfps list, as VCL_StackVFP does, rather than the vdps. Add a test for a VMOD-defined VDP.
-
Nils Goroll authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
-
Nils Goroll authored
... which would require error handling on all of the call stack upwards, which is not that trivial an endavour. Also, for the calls via VSMW_DestroyCluster(), we are actually about to free space, but temporarily need more. Other ideas: - panic, but leave a better error message - monitor free space and refuse loading vcls if insufficient - maybe as part of some more general vcl management (e.g. "maintain max xx vcls loaded - aka auto-discard) Ref #2764
-
- 17 Apr, 2019 5 commits
-
-
Nils Goroll authored
-
Dridi Boukelmoune authored
-
Nils Goroll authored
We only need to wake up vbp_thread if the probe we insert on the binheap has become the next due. As a binheap_root() call is cheap and saves posting many signals, this optimization should be worth some extra lines. Also we add a function for the two places where we insert into the heap. Ref #2976
-
Nils Goroll authored
We could also move the VSL() outside if we copied dir->vcl_name
-
Nils Goroll authored
The due time of the probe which has just finished may be earlier than the previously earliest due time, so vbp_thread needs to be woken up to update the nxt value Fixes #2976
-
- 16 Apr, 2019 5 commits
-
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
-
Federico G. Schwindt authored
Via lgtm.
-
Federico G. Schwindt authored
-
Dridi Boukelmoune authored
This reverts commit e7d5fb68. Apparently this change looks intentional, but it should be revisited as it makes troubleshooting significantly harder when logs (for example Backend_health records) aren't explicit about the backend they are referring to.
-