#!/bin/bash # tmp="/tmp/`basename $0`-$$" tmp0="$tmp.0.sql" tmp1="$tmp.1.sql" tmp2="$tmp.2.sql" tmp3="$tmp.3.sql" tmp4="$tmp.4.sql" msdd="1,000,000,000" msd="`echo "$msdd" | tr -d ,`" # usAge() { echo "usAge: `basename $0` [-s sdiffX] [w] 0|[b startblock]|[s shift] endblock|z" echo " run as root or change dosql1()" echo echo " -s sdiffX = all shares must have minsdiff <= sdiffX - if not then abort" echo " any commas are removed with \"tr -d ,\"" echo " shares with sdiff below sdiffX are ignored. Default $msdd" echo " w = worker stats - without w = user stats" echo " 0 = from the first hi share" echo " b startblock = start after the workinfoid of startblock (instead of 0)" echo " i.e. this only includes data AFTER 'startblock'" echo " 'startblock' is the lowest block >= 'startblock'" echo " s shift = start from the shift with the starting 5 char code 'shift'" echo " if there ever is a duplicate, it will use the lowest one" echo " endblock = finish at the workinfoid of endblock (instead of z)" echo " this includes all of 'endblock'" echo " 'endblock' is the highest block <= 'endblock'" echo " z = to the last high share" echo echo " Basically would be '0 z' for all available share history or" echo " e.g. 'b NNN MMM' for a block range after NNN up to (including) MMM or" echo " e.g. 'b NNN z' starting after block NNN up to now" echo " Put a 'w' in front, to get worker instead of user summarisation" echo echo " The share history determines the limits used" echo " i.e. markersummaries before the first share, or after the last share," echo " are not included since that would invalidate the results" echo " Missing shares within the share data would invalidate the results" exit 1 } # dosql1() { su - postgres << EOF echo "$sql" | psql ckdb EOF } # procsql() { got="" fin="" prev="" while true ; do read line if [ "$?" != "0" ] ; then break fi if [ "$got" ] ; then if [ -z "$fin" ] ; then echo "$line" if [ "${line:0:1}" = "(" ] ;then fin="y" fi fi else if [ "${line:0:5}" = "-----" ] ; then got="y" echo "$prev" echo "$line" else prev="$line" fi fi done } # dosql() { dosql1 2>&1 | procsql } # # reduce a large, 9 or more digit, comma number with 2 decimal places to nnG toG() { sed -e "s/,\([0-9][0-9]\)[0-9],[0-9][0-9][0-9],[0-9][0-9][0-9]\.[0-9][0-9]*/.\1G/g" -e "s/\(\.[0-9][0-9]G\) *|/\1 |/g" -e "s/ *$//" } # dedup() { cut -d '|' -f 1-5,7-10,12 } # dedupw() { cut -d '|' -f 1-4,6-9,11 } # p1="$((0x1d))" p1v="`echo "$p1 3 - 8 * p" | dc`" # # calculate the diff ratio for each diffacc, given bits diff1() { read line read line while true ; do read line if [ "$?" != "0" ] ; then break fi if [ "${line:0:1}" = "(" ] ; then break fi bits="`echo "$line" | cut -d '|' -f2 | tr -d ' '`" diffacc="`echo "$line" | cut -d '|' -f3 | tr -d ' '`" po="$((0x${bits:0:2}))" bi="$((0x${bits:2:6}))" pd="`echo "$p1v $po 3 - 8 * - p" | dc`" if [ "$pd" -lt "8" ] ; then pd="8" fi per="`echo "8 k $diffacc 2 $pd ^ 65535 * $bi / / p" | dc`" echo "$line | ${per}" done } # diffvals() { grep "^$uorw *|" "$tmp4" | cut -d '|' -f4 | tr "\n" "+" | sed -e "s/+/ +/g" } # getdiff() { echo "4 k 0 `diffvals` p" | dc } # # blocks are appended 3rd since if blocks exist, then at least 1 share must proc() { read line read line while true ; do read line if [ "$?" != "0" ] ; then break fi if [ "${line:0:1}" = "(" ] ; then echo "$line" break fi uorw="`echo "$line" | cut -d'|' -f1 | tr -d ' '`" diffs="`getdiff` BDR" if [ "${diffs:0:1}" = "." ] ; then diffs=" 0$diffs" elif [ "${diffs:1:1}" = "." ] ; then diffs=" $diffs" elif [ "${diffs:2:1}" = "." ] ; then diffs=" $diffs" fi stats="`grep "^$uorw *|" "$tmp2"`" bstats="`grep "^$uorw *|" "$tmp3"`" if [ "$stats" ] ; then s=" | " else s="" fi if [ "$bstats" ] ; then b=" | " else b="" fi echo "$line | $diffs$s$stats$b$bstats" done } # procblk() { while true ; do read line if [ "$?" != "0" ] ; then break fi if [ "${line:0:1}" = "(" ] ; then echo "$line" break fi uorw="`echo "$line" | cut -d'|' -f1 | tr -d ' '`" bstats="`grep "^$uorw *|" "$tmp3"`" echo "$line | $bstats" done } # if [ "$1" = "-s" ] ; then shift msd="`echo "$1" | tr -d ,`" shift fi # if [ -z "$1" ] ; then usAge fi # if [ "$1" = "-?" -o "$1" = "-h" -o "$1" = "-help" -o "$1" = "--help" ] ; then usAge fi # work="" if [ "$1" = "w" ] ; then work="w" shift fi # valsh="sdiff>=$msd" # Orphans included, Rejects not - since they can be below diff, or invalid, or too late or ... valblk="expirydate>'5555-05-05' and confirmed!='R'" # # START if [ "$1" = "0" ] ; then shift selwm1="ms.markerid>=(select min(markerid) from workmarkers where workinfoidstart>=(select min(workinfoid) from shares))" selsh1="" selblk1="and b.workinfoid>=(select min(workinfoid) from shares)" else if [ "$1" = "s" ] ; then shift selwm1="ms.markerid>=(select min(markerid) from workmarkers where description like 'Shift fin: $1 %')" selsh1="and workinfoid>=(select min(workinfoidstart) from workmarkers where description like 'Shift fin: $1 %')" selblk1="$selsh1" shift else if [ "$1" = "b" ] ; then shift selwm1="ms.markerid>=(select min(markerid) from workmarkers where workinfoidstart>(select min(workinfoid) from blocks where height>=$1))" selsh1="and s.workinfoid>=(select min(workinfoid) from blocks where height>=$1)" selblk1="and b.height>$1" shift else echo "ERR: Unknown first option '$1'" usAge fi fi fi # # END if [ "$1" = "z" ] ; then selwm2="" selsh2="" selblk2="" else if [ -z "$1" ] ; then echo "ERR: missing ending block|z" usAge else selwm2="and wm.markerid<=(select max(markerid) from workmarkers where workinfoidend<=(select max(workinfoid) from blocks where height<=$1))" selsh2="and s.workinfoid<=(select max(workinfoid) from blocks where height<=$1)" selblk2="and b.height<=$1" fi fi # # check all shares have minsdiff > $msd sql="select count(*) from shares s where s.minsdiff>$msd $selsh1 $selsh2 ;" # dosql > "$tmp0" count="`head -n 3 "$tmp0" | tail -n 1`" if [ "$count" != "0" ] ; then count echo "ERROR: there were '$count' shares with minsdiff>$msd - that must be '0'" usAge fi # # summarise markersummary into tmp1 for any user or worker with 'msd' diffacc or more if [ -z "$work" ] ; then sql="select ms.userid,substring(max(ms.workername) from '(^[^\._]*)'),to_char(sum(ms.diffacc),'999G999G999G999G999.99'),to_char(sum(ms.sharecount),'999G999G999G999G999') from markersummary ms where $selwm1 $selwm2 group by ms.userid having sum(ms.diffacc)>$msd order by sum(ms.diffacc) desc;" else sql="select ms.workername,to_char(sum(ms.diffacc),'999G999G999G999G999.99'),to_char(sum(ms.sharecount),'999G999G999G999G999.99') from markersummary ms where $selwm1 $selwm2 group by ms.workername having sum(ms.diffacc)>$msd order by sum(ms.diffacc) desc;" fi # dosql > "$tmp1" # # summarise the shares into tmp2 if [ -z "$work" ] ; then sql="select s.userid,count(*),to_char(min(s.sdiff),'999G999G999G999G999.99'),to_char(max(s.sdiff),'999G999G999G999G999.99'),to_char(avg(s.sdiff),'999G999G999G999G999.99') from shares s where $valsh $selsh1 $selsh2 group by s.userid;" else sql="select s.workername,count(*),to_char(min(s.sdiff),'999G999G999G999G999.99'),to_char(max(s.sdiff),'999G999G999G999G999.99'),to_char(avg(s.sdiff),'999G999G999G999G999.99') from shares s where $valsh $selsh1 $selsh2 group by s.workername;" fi # dosql > "$tmp2" # # summarise the blocks into tmp3 if [ -z "$work" ] ; then sql="select b.userid,count(*)||' B' from blocks b where $valblk $selblk1 $selblk2 group by b.userid;" else sql="select b.workername,count(*)||' B' from blocks b where $valblk $selblk1 $selblk2 group by b.workername;" fi # dosql > "$tmp3" # # summarise diff% into tmp4 (all ids will have a value) if [ -z "$work" ] ; then sql0="select ms.userid,wi.bits,sum(ms.diffacc)" sqlz="group by ms.userid,wi.bits;" else sql0="select ms.workername,wi.bits,sum(ms.diffacc)" sqlz="group by ms.userid,ms.workername,wi.bits;" fi # valwm="wm.status='p' and wm.expirydate>'5555-05-05'" sql="$sql0 from markersummary ms join workmarkers wm on ms.markerid=wm.markerid join workinfo wi on wm.workinfoidend=wi.workinfoid where $valwm and $selwm1 $selwm2 $sqlz" # dosql | diff1 > "$tmp4" # # join (append) to each tmp1 row, the matching row/data in tmp4, tmp2 and tmp3 cat "$tmp1" | proc | toG | dedup$work # shred -uz "$tmp0" "$tmp1" "$tmp2" "$tmp3" "$tmp4" &> /dev/null