willy tarreau | 1f431b5 | 2006-05-21 14:46:15 +0200 | [diff] [blame] | 1 | '+' = done, '-' = todo, '*' = done except doc |
willy tarreau | 814cbc6 | 2006-05-13 13:00:29 +0200 | [diff] [blame] | 2 | |
| 3 | 1.2.12 : |
| 4 | + weighted RR/SH |
| 5 | |
| 6 | 1.2.13 : |
| 7 | + maxconn |
| 8 | + queueing |
| 9 | |
| 10 | 1.2.14 : |
willy tarreau | 1f431b5 | 2006-05-21 14:46:15 +0200 | [diff] [blame] | 11 | * HTML status page |
| 12 | |
| 13 | stats enable |
| 14 | stats uri /?stats |
| 15 | stats realm w.ods.org\ statistics |
| 16 | stats auth user1:pass1 |
| 17 | stats auth user2:pass2 |
| 18 | stats auth user3:pass3 |
| 19 | stats scope <px_id> | '.' |
willy tarreau | 814cbc6 | 2006-05-13 13:00:29 +0200 | [diff] [blame] | 20 | |
willy tarreau | 38d7906 | 2006-05-21 14:47:13 +0200 | [diff] [blame] | 21 | + allow server-less proxies (for stats) |
| 22 | |
willy tarreau | 814cbc6 | 2006-05-13 13:00:29 +0200 | [diff] [blame] | 23 | - separate timeout controls |
| 24 | |
| 25 | - option 'abortonclose' : if the session is queued or being connecting |
| 26 | to the server, and the client sends a shutdown(), then decide to abort |
| 27 | the session early because in most situations, this will be caused by |
| 28 | a client hitting the 'Stop' button, so there's no reason to overload |
| 29 | the servers with unservable requests. However, this is not HTTP compliant |
| 30 | and might cause little trouble to some very specific clients used to |
| 31 | close immediately after sending the request (no support for KA, which ones?) |
| 32 | |
willy tarreau | f76e6ca | 2006-05-21 21:09:55 +0200 | [diff] [blame^] | 33 | + minconn : makes the server's maxconn dynamic, which will be computed as a |
willy tarreau | 814cbc6 | 2006-05-13 13:00:29 +0200 | [diff] [blame] | 34 | ratio of the proxy's sessions : |
| 35 | srv->effective_maxconn = |
| 36 | max(srv->maxconn * px->nbsess / px->maxconn, srv->minconn) |
| 37 | |
willy tarreau | 814cbc6 | 2006-05-13 13:00:29 +0200 | [diff] [blame] | 38 | 1.3 : |
| 39 | - handle half-closed connections better (cli/srv would not distinguish |
| 40 | DATA/SHUTR/SHUTW, it would be a session flag which would tell shutr/shutw). |
| 41 | Check how it got changed in httpterm. |
| 42 | |
| 43 | - 3 memory models : failsafe (prealloc), normal (current), optimal (alloc on |
| 44 | demand) |
| 45 | |
| 46 | - wait queues replaced for priority-based trees |
| 47 | |
| 48 | - ability to assign a prio based on L7 matching |
| 49 | |
| 50 | - prio-based O(1) scheduler |
| 51 | |
| 52 | - maxconn reserve for VIP/admins |
| 53 | |
| 54 | - verify if it would be worth implementing an epoll_ctl_batch() for Linux |
| 55 | |