Driver-Taper protocol

From wiki.zmanda.com
Revision as of 17:39, 18 January 2010 by Dustin (talk | contribs) (→‎Command sequence during backup operation: represent newest version of protocol - no DIRECTTCP-PORT command)
Jump to navigation Jump to search

Communication method

driver talks via two pipes connected to taper's stdin and stdout. The commands and responses are plain text.

Command sequence during backup operation

  • driver initialises taper with: START-TAPER <timestamp> to which taper replies with: TAPE-OK or, for fatal errors, with: TAPE-ERROR [<message>]. For speed up,taper is authorized to scan the library and find a valid tape after it received the START-TAPER command.
  • driver can ask taper to copy a file from the holding disk to tape (FILE-WRITE) or directly from a dumper (PORT-WRITE) or exit at the end of the run (QUIT).
  • taper responds to the PORT-WRITE command with:
    • PORT <hdr-port> <ip-port-pairs> which the driver should then hand on to dumper in a PORT-DUMP command
  • if taper has no tape in use, it reply with REQUEST-NEW-TAPE.
  • If the copy to tape finishes correctly taper replies with DONE.
  • If something goes wrong with the tape, taper can ask to continue on a new tape REQUEST-NEW-TAPE (for splited dump only) or it can abort that dump with PARTIAL or FAILED reply
  • After any dump that finished in PARTIAL or FAILED with a TAPE-ERROR, the driver must send a NEW-TAPE before issuing another *-WRITE command.

Other commands

  • If driver says something that taper doesn't recognise it responds with: BAD-COMMAND <message>
  • taper responds to the QUIT command with: QUITING

Protocol command reference

Taper command reply

Reply Description
TAPE-OK taper is setup correctly
TAPE-ERROR Error in setup of the driver
PORT Reply sent in response to PORT-WRITE command for AMANDA datapath
DIRECTTCP-PORT Reply sent in response to PORT-WRITE command for DIRECTTCP datapath
DONE Full dump is on the media
PARTIAL Dump was partially written to the media.
FAILED Nothing was written to the media
DUMPER-STATUS Ask driver to send the dumper status
PARTDONE A complete part is written to tape
REQUEST-NEW-TAPE Ask driver if it should use a new tape
NEW-TAPE Will continue on a new found tape
NO-NEW-TAPE Will not continue on a new tape


DONE/PARTIAL and FAILED reply also contains an INPUT-* an TAPE-* message to denote any error:

Reply Description
INPUT-GOOD There was no error with the input.
INPUT-ERROR There was error with the input.
Reply Description
TAPE-GOOD This tape can be use to write another dump.
TAPE-ERROR Nothing more can fit on that tape.

A result for a successful dump should be: DONE INPUT-GOOD TAPE-GOOD

Driver/Taper Requests/Replies

Driver request Taper reply Description
initiall setup
START-TAPER --->
<--- TAPER-OK normal taper setup (a tape is available, nothing is written on it)
<--- TAPE-ERROR failed taper setup (no tape are available or something else is broken)
file-write setup
FILE-WRITE --->
<--- REQUEST-NEW-TAPE (Optional) ask driver if we should start this dump on a new tape. (driver will send a NEW-TAPE or NO-NEW-TAPE command)
Will get one part-result for each part written and one global result
port-write setup
PORT-WRITE --->
<--- PORT
<--- REQUEST-NEW-TAPE (Optional) ask driver if we should start this dump on a new tape. (driver will send a NEW-TAPE or NO-NEW-TAPE command)
Will continue with one part-result for each part written and one global result
global-result
Can ask-dumper-status (Should be done if taper detect no error (INPUT-GOOD and TAPE-GOOD)
<--- DONE INPUT-GOOD TAPE-GOOD normal protocol for a success
<--- PARTIAL INPUT-* TAPE-* protocol for error in data phase (something written to tape)
<--- FAILED INPUT-* TAPE-* protocol for error in setup (before something is written to tape)
ask-dumper-status
<--- DUMPER-STATUS Ask driver the dumper status
DONE ---> Dumper succeeded
FAILED ---> Dumper failed
part-result
<--- PARTDONE successfully part written to tape
<--- REQUEST-NEW-TAPE We get a tape error, ask the driver if we can use a new tape (driver will send a NEW-TAPE or NO-NEW-TAPE command)
driver don't want the taper to use a new tape
NO-NEW-TAPE ---> driver doesn't want the taper to use a new tape (continue with global-result)
driver want the taper to use a new tape
NEW-TAPE ---> driver want the taper to use a new tape
<--- NEW-TAPE taper found one
<--- NO-NEW-TAPE no tape are available
quit
QUIT ---> taper quits

Command/reply arguments

Protocol data

data description
<timestamp> Time as "yymmdd" of "yymmddhhmmss"
<handle> Request ID
<filename> Name of file on the holding disk where backup will be written to
<port> Taper port to send the backup data to
<host> Hostname of the client
<disk> Disk on the client being backed up
<level> Dump level being used for backup
splitsize size of each part on tape in bytes
split_diskbuffer directory in which to create disk buffer files, or the literal string "NULL" if none is specified
<message> Error or Status message

driver command

  • START-TAPER timestamp
  • PORT-WRITE handle hostname diskname level datestamp splitsize split_diskbuffer fallback_splitsize datapath
    • if there is no split_diskbuffer, then it is specified as the string "NULL"
    • splitsize and fallback_splitsize, both in kb, come directly from the dumptype
  • FILE-WRITE handle filename hostname diskname level datestamp splitsize
  • DONE handle (result from dumper send to taper)
  • FAILED handle (result from dumper send to taper)
  • NEW-TAPE
  • NO-NEW-TAPE "reason"
  • QUIT

handles are unique to each DLE during a run, but since the taper only operates on a single file at a time, they are generally useless to the taper except in constructing syntactically correct responses to the driver.

taper reply

  • TAPER-OK
  • TAPE-ERROR handle "error-message"
  • PARTIAL handle INPUT-* TAPE-* "[sec %f kb %d kps %f]" "input-error-message" "tape-error-message"
    • The statistic are the sum of all successful part (PARTDONE), it should not include the failed part.
  • DONE handle INPUT-GOOD TAPE-GOOD "[sec %f kb %d kps %f]" "" ""
    • The statistic are the sum of all successful part (PARTDONE), it should not include the failed part.
  • FAILED handle INPUT-* TAPE-* "input-error-message" "tape-error-message"
  • NEW-TAPE handle label
  • NO-NEW-TAPE handle
  • PARTDONE handle label fileno kb "[sec %f kb %d kps %f]"
    • yep, the kb are specified twice
  • REQUEST-NEW-TAPE handle
  • PORT port
  • DIRECTTCP-PORT hdr-port ip-port-pairs
  • BAD-COMMAND "error message"
  • DUMPER-STATUS handle

LOG

  • One PART or PARTPARTIAL log line for all part written to tape. It tell the location and status of each part.
    • PART taper label fileno hostname diskname timestamp part-number/totpart level [sec %f kb %d kps %f]
    • PARTPARTIAL taper label fileno hostname diskname timestamp part-number/totpart level [sec %f kb %d kps %f] "error message"

totpart should be -1 if the the number is not known (PORT-WRITE)

  • One DONE/PARTIAL/FAILED for each dump. It tell the status of the complete dump.
    • DONE taper hostname diskname timestamp totpart level [sec %f kb %d kps %f]
      • The statistic is the sum of all successful part (PART), it doesn't include PARTPARTIAL
      • This indicates that, from the taper's perspective, the dump is completely done and should not be retried; this is the case even if the dumped file was partial
    • PARTIAL taper hostname diskname timestamp totpart level [sec %f kb %d kps %f] "error message"
      • The statistic is the sum of all successful part (PART), it doesn't include PARTPARTIAL
      • Some of the data was sent to the device, but not all of it; the driver will likely retry the dump
    • FAILED taper hostname diskname timestamp level "error message"
      • None of the data was sent to the device; the driver will likely retry the dump

algorithm after a part written to tape

 if (successful part) {
   L_PART -> logfile
   if (splited dump) {
     PARTDONE -> DRIVER
   if (not splited dump || last part) {
     if (PORT-WRITE) {
       DUMPER-STATUS -> driver
       get dumper result from driver (FAILED|DONE)
     }
     if (partial) #from dumper status or partial flag in header {
       PARTIAL INPUT-GOOD TAPE-GOOD -> driver
       L_PARTIAL -> logfile
     } else (whole) {
       DONE INPUT-GOOD TAPE-GOOD -> driver
       L_SUCCESS -> logfile
     }
   } else #splited dump
     GOTO next part
 } else #failed part {
   if (something written to tape for this part)
     L_PARTPARTIAL -> logfile
   if (splitted dump and tape error && no-input-error) {
     SPLIT_NEEDNEXT -> driver
     L_INFO -> logfile
     #wait for driver NEW-TAPE or NO-NEW-TAPE
     #It could take a long time to get the result
     if(driver sent NEW-TAPE)
       if (new tape available)
         NEW-TAPE -> DRIVER
         GOTO retry that part to the new tape
       else #no valid writable tape found
         NO-NEW-TAPE -> DRIVER
     else #driver sent NO-NEW-TAPE
       noop
   }
   #it's possible to have both INPUT-* and TAPE-* error or none.
   if (nothing written to tape for this dump) {
     FAILED INPUT-* TAPE-*  -> driver
     L_FAIL -> logfile
   } else #something written to tape for this dump {
     PARTIAL INPUT-* TAPE-* -> driver;
     L_PARTIAL -> logfile
   }
 }