#!/bin/bash

# Please see this shell script rather as template how to call rqg_batch.pl even though
# it might be already in its current state sufficient for doing a lot around RQG.
#

USAGE="USAGE: $0 <Config file> <Basedir>"
CALL_LINE="$0 $*"

set -e

# Config file for rqg_batch.pl containing various settings for the RQG+server+InnoDB etc.
# including settings for avoiding open bugs.
CONFIG="$1"
if [ ! -e "$CONFIG" ]
then
   echo "The config file '$CONFIG' does not exist."
   echo "The call was ->$CALL_LINE<-"
   echo $USAGE
   exit
fi

# Path to MariaDB binaries
BASEDIR="$2"
if [ "$BASEDIR" = "" ]
then
   echo "You need to assign a basedir == path to MariaDB binaries like '/work/10.3/bld_asan'"
   echo "The call was ->$CALL_LINE<-"
   echo $USAGE
   exit
fi
if [ ! -d "$BASEDIR" ]
then
   echo "BASEDIR '$BASEDIR' does not exist."
   exit
fi

CASE=`basename $CONFIG .cfg`
if [ $CASE = $CONFIG ]
then
   CASE=`basename $CONFIG .cc`
fi

PROT="$CASE"".prt"

# My standard work directory for rqg_batch.pl.
# The workdirs for ongoing RQG runs are in its sub directories.
# The BATCH_WORKDIR is also for permanent storing results and similar.
BATCH_WORKDIR="storage"
if [ ! -d "$BATCH_WORKDIR" ]
then
   mkdir $BATCH_WORKDIR
fi

# My standard var directory on tmpfs (recommended) for rqg_batch.pl.
# The vardirs for ongoing RQG runs are in its sub directories.
# rqg_batch.pl cleans BATCH_VARDIR except it gets killed or dies because of internal error.
BATCH_VARDIR="/dev/shm/vardir"
if [ ! -d "$BATCH_VARDIR" ]
then
   mkdir $BATCH_VARDIR
fi

set +e

# Go with heavy load in case the rqg_batch.pl ResourceControl allows it.
# The rqg_batch.pl ResourceControl should be capable to avoid trouble with resources.
# Per experience:
# More general load on the testing raises the likelihood to find or replay a
# concurrency bug.
PARALLEL=`nproc`
PARALLEL=$(($PARALLEL * 3))


# TRIALS is used as
# - one of two limits (TRIALS and MAX_RUNTIME) for the size of a testing campaign
#   Whenever one of these gets exceeded rqg_batch.pl stops ongoing RQG runs,
#   makes a cleanup, gives a summary and exits.
#   In case of
#   - (mostly) hitting no internal error in the RQG runner ('rqg.pl'), the RQG core (lib/*) and
#     ingredients invoked (reporter, validator, grammar, ...)
#     MAX_RUNTIME is the better limiter
#   - (sometimes) hitting some internal error in ... (see above)
#     TRIALS is here the better limiter because the situation is roughly hopeless and
#     you get an earlier end with less resource consumption.
# - used for limiting certain phases during RQG test simplification if running that at all.
# TRIALS means regular finished (!= stopped by rqg_batch.pl because of whatever reason) RQG runs.
TRIALS=$(($PARALLEL * 10))

# MAX_RUNTIME is a better limit than TRIALS for defining the size of a testing campaign
# when running 'production' == QA.
# Please be aware that the runtime of util/issue_grep.sh is not included.
# RQG batch run elapsed runtime =
#    assigned max_runtime
# +  time for stopping the active RQG Worker (less than 3 seconds)
# +  util/issue_grep.sh elapsed time =
#       no of logs in last_batch_workdir * (1 till 3 seconds depending on log size)
MAX_RUNTIME=1800


# Only one temporary 'God' (rqg_batch.pl vs. concurrent MTR, single RQG or whatever) on testing box
# -------------------------------------------------------------------------------------------------
# in order to avoid "ill" runs where
# - current rqg_batch run ---- other ongoing rqg_batch run
# - current rqg_batch run ---- ongoing MTR run
# clash on the same resources (vardir, ports -> MTR_BUILD_THREAD, maybe even files) or
# suffer from tmpfs full etc.
killall -9 perl ; killall -9 mysqld
rm -rf /dev/shm/var*

# In case the cleanup above is disabled than at least this.
rm -rf $BATCH_VARDIR/*

# 1. There should be usually sufficient space in VARDIR for just a few fat core files caused
#    by ASAN.
#    Already the RQG runner will take care that everything important inside his VARDIR will be
#    saved in his WORKDIR and empty his VARDIR. rqg_batch.pl will empty the VARDIR of this RQG
#    runner again. So the space comsumption of a core is only temporary.
#    The rqg_batch.pl ResourceControl will also take care to avoid VARDIR full.
# 2. If its not an ASAN build than this environment variable is harmless anyway.
export ASAN_OPTIONS=abort_on_error=1,disable_coredump=0
echo "Have set "`env | grep ASAN`

rm -f $PROT


set -o pipefail
# Remove the logs of RQG runs achieving STATUS_OK/verdict 'ignore_*'.
# Their stuff grammar/datadir was not archived and is already thrown away.
# So basically:
# Remove the --discard_logs in case you want to see logs of RQG runs which achieved
# the verdict 'ignore_*' (blacklist match or STATUS_OK or stopped by rqg_batch.pl)
# --discard_logs                                                         \
#
# Do not abort if hitting Perl errors or STATUS_ENVIRONMENT_FAILURE. IMHO rather questionable
# --force
#
# Old stuff. In history required.
# --no-mask  Unclear if ./rqg_batch.pl and rqg.pl will mask at all.
#
# rqg_batch.pl prints how it would start RQG Workers and the RQG Worker started "fakes" that
# it has achieved the verdict assigned. == There all no "real" RQG runs at all.
# Example:
# --dryrun=replay --> All RQG Worker started "tell" that they have achieved some replay.
# Use cases:
# a) When using the Combinator see which combinations would get generated.
# b) When using the Simplifier see how it would be tried to shrink the grammar.
# c) --dryrun=ignore_blacklist see how TRIALS would be the limiter.
# --dryrun=ignore_blacklist                                            \
# --dryrun=replay                                                      \
#
# rqg_batch stops immediate if hitting a 'replay'
# --stop_on_replay                                                     \
#
# Use this grammar in all test variants
  GRAMMAR=MDEV-20399.yy
  vi $GRAMMAR

# In case you distrust the rqg_batch.pl config file etc. going with some limited number is useful.
# TRIALS=2
# PARALLEL=2
# TRIALS=1
# PARALLEL=1

nohup perl -w ./rqg_batch.pl                                           \
--workdir=$BATCH_WORKDIR                                               \
--vardir=$BATCH_VARDIR                                                 \
--parallel=$PARALLEL                                                   \
--basedir1=$BASEDIR                                                    \
--config=$CONFIG                                                       \
--trials=$TRIALS                                                       \
--max_runtime=$MAX_RUNTIME                                             \
--discard_logs                                                         \
  --stop_on_replay                                                     \
  --grammar=$GRAMMAR                                                   \
  --threads=2                                                          \
--no-mask                                                              \
--type=RQG_Simplifier                                                  \
--script_debug=_nix_                                                   \
> $PROT 2>&1 &

# Avoid that "tail -f ..." starts before the file exists.
sleep 2
tail -f $PROT