Squid-3.0: Feature freeze: - overlapping helper feature implementation (plain or stateful or both). Other? - Windows port (Merging in process, won't delay 3.0). - Generalize socket binding to allow for multipe ICP/HTCP/SNMP sockets (get rid of udp_incoming_address) (???) Done items: - epoll linkage. - reply_body_max_size syntax cleanups. (done) - Other sponsored work that needs to get into mainline sooner rather than later (None known of - RC). - Update automake and autoconf on squid-cache.org for the squid-3 CVS snapshot generation. (done) - ESI client side streams support framework (done). - ESI (done). - Authentication tidyups in proxy_auth acl managagement (Done). - Appropriate parts of rproxy (Done). - Some kind of Range support to at least the level of 2.5 and earlier (Done). - Someone other than Robert to review the draft squid 3 release notes (in CVS already). Squid-3.1 - always/never direct syntax cleanups. - refactoring of acl driven types to reduce amount of duplicated code (acl_check, acl_tos, acl_address, acl_size_t, ...) - Go over the old TODO (below) and sort out what has o) Been done o) Needs to be done, and when o) Can be ignored - ETag caching (???) - Customized logging capabilities - External logging - IPv6 - Other devel.squid-cache.org projects - header filtering improvements (bug #601) Later: - Rework the store API like planned - Improved event driven comm code - Buffer management (required by store API and comm code improvements) - Improved reply processing to not reparse the reply headers several times - Implement HTCP authentication - Log HTCP queries somewhere, like htcp_query.log - There are various places where ICP means ICP/HTCP, and other places where it does not. For example, cachemgr 'counters' stats like icp.pkts_sent count only ICP. However, in 'peer_select' stats, ICP does include HTCP as well. ****************************************************************************** Our OLD Todo List, still containing some interesting items. ****************************************************************************** () dont re-sibling requests which came from a sibling () (1) is high priority, (5) is low priority (**) Means we need this for the real 1.2 release ============================================================================== (1) A customizable cache replacement policy. Ugh, this could be interesting since we just optimized the LRU replacement with a doubly-linked list. (2) be more persistent in forwarding requests; return fewer error pages (2) Fix some problems with peer-selection. - Allow for 'backup' parents - An orderd list of siblings to try. - Hash function load balancing (2) Calculate Content-Length for multipart range replies (AR) (3) hit metering (3) ETag support (makes metadata bigger--ack!) (3) Efficient public peer access control *without ACLs* Bill Wichers Dave Zarzycki # distant_peer_deny (icmp rtt) (hops) # If the ICP client is farther than "z" hops away, # or if more than y% of pings to the client exceed # x milliseconds, then deny. distant_peer_deny 200 75 10 (3) deny/allow tags for HTTP header fields (AR) (3) LDAP support. See http://miso.wwa.com/~donley/ and Clayton Donley (3) What to do about ACL's and URL escaping? (3) Review RFC 2068 and RFC 2109 for header handing, especially Cache-Control. (4) ** Everywhere that we use 'pattern' or such, use ACL elements instead. stoplist_pattern, refresh_pattern, ... (DW) (4) Refresh based on content types. This means we'll need an enum of known content types added to StoreEntry. Unknown types will lose. (4) Write binary headers as metadata? (4) HTML-escape special characters in errorConvert(). (5) X-Proxy-hops header? (5) internalize redirectors? (5) For _popular_ objects, return hits if known to be stale, but then initiate an IMS/refresh in the background. (5) #ifdefs to disable IP caching (5) REST for failed ftp transfers. (5) Customizable access logging. (5) Inline DNS queries with native resolver calls. (5) MD5 acl type (5) return "ERR_TOO_LARGE" messages for replies with content-lengths larger than some limit. (5) Bind HTTP/ICP sockets to multiple interfaces (5) Custom X-Forwarded-For DONE: + s/sprintf/snprintf/g + Write the text for all the error message files + fix all the assert(!ERR_FOO) placeholders + Finish hacking at storeCopy() to achieve in-memory sliding window for objects being retrieved. + make USE_PROXY_AUTH standard + single linked list for disk LRU replacement + Incorporate Pelletier's latest retry patch for 1.1 (3) URLs --> MD5? (1) reconfigure (SIGHUP) causes problems with Config.cacheSwap.swapDirs[]. (1) I think we're assuming persistent connections and not looking at the reply headers. (2) Write src/store_key_md5.c and lib/md5.c (2) SNMP interface (5) convert ipcache/fqdncache to double-linked list and lose qsort() calls. (3) AS # routing - whois -h whois.ra.net '\!gAS194' - Stevens V2 Ch.18 (3) SIGHUP should reread error text files (4) support ftp://host/pub;type=d (5) should we write clean store logs from the double-linked list so that they are sorted in order of last reference? Then upon restart we won't have to worry about the list being unsorted and causing problems with the LRU magic value. (3) ** Make checkFailureRatio() work again. (DW) (2) ** copy over -z semantics from 1.1; don't automatically wipeout cache (DW) (3) ** Make swap.state binary with fixed-size cache keys (AR) (3) ** Make sure every squid.conf option has documentation (??) (3) ** Store cache key and other metadata in swap files (KA) (3) ** Finish implementing the *_dump() functions in cache_cf.c (DW) (2) ** Fix cachemgr.cgi. It doesn't understand some of the new output formats. (??) (2) ** Add "Squid" trailer to all error messages. (AR) (2) ** copy over 'test_reachability' from 1.1 (DW) (2) ** we don't properly handle the HTTP/1.1 'Connection' header (KA) (2) ** Code cleanup: (KA) - Make sure the lib/*.c files do not use printf/fprintf. - Replace bcopy with xmemcpy (4) ** FTP PUT (KA) (1) SSL doesn't work with Proxy Authentication (3) HTTP Header todo list (AR) (5) replace "for" loops in md5.c with memcpy and memset (3) ** shrink-able memory pools (AR) (3) ** request-range (AR) (4) ** Double check that MemBuf.size is used correctly everywhere (AR) (3) 'no_cache' access list for specifing objects which should not be cached. (3) Full request headers for CONNECT requests. (1) ** Send full HTTP requests in ICP messages (DW) (2) Track cache keys (MD5) in memory utilization (3) Resource limits. (3) Arjan's external proxy auth patch is at http://www.iae.nl/users/devet/squid/proxy_auth/ (3) In some situations, a private-key reply needs to eject a public-key cached object. For example when an object is first cachable, but later becomes uncachable. (4) Configurable proxy-auth realm (4) Configurable FQDN cache size