# See the file LICENSE for redistribution information. # # Copyright (c) 2003, 2012 Oracle and/or its affiliates. All rights reserved. # # $Id$ # # TEST rep017 # TEST Concurrency with checkpoints. # TEST # TEST Verify that we achieve concurrency in the presence of checkpoints. # TEST Here are the checks that we wish to make: # TEST While dbenv1 is handling the checkpoint record: # TEST Subsequent in-order log records are accepted. # TEST Accepted PERM log records get NOTPERM # TEST A subsequent checkpoint gets NOTPERM # TEST After checkpoint completes, next txn returns PERM proc rep017 { method { niter 10 } { tnum "017" } args } { source ./include.tcl global repfiles_in_memory # Run for all access methods. if { $checking_valid_methods } { return "ALL" } set args [convert_args $method $args] set logsets [create_logsets 2] set msg2 "and on-disk replication files" if { $repfiles_in_memory } { set msg2 "and in-memory replication files" } # Run the body of the test with and without recovery. foreach r $test_recopts { foreach l $logsets { set logindex [lsearch -exact $l "in-memory"] if { $r == "-recover" && $logindex != -1 } { puts "Rep$tnum: Skipping\ for in-memory logs with -recover." continue } puts "Rep$tnum ($method $r):\ Concurrency with checkpoints $msg2." puts "Rep$tnum: Master logs are [lindex $l 0]" puts "Rep$tnum: Client logs are [lindex $l 1]" rep017_sub $method $niter $tnum $l $r $args } } } proc rep017_sub { method niter tnum logset recargs largs } { source ./include.tcl global perm_response_list global repfiles_in_memory global rep_verbose global verbose_type set verbargs "" if { $rep_verbose == 1 } { set verbargs " -verbose {$verbose_type on} " } set repmemargs "" if { $repfiles_in_memory } { set repmemargs "-rep_inmem_files " } env_cleanup $testdir set omethod [convert_method $method] replsetup $testdir/MSGQUEUEDIR set masterdir $testdir/MASTERDIR set clientdir $testdir/CLIENTDIR file mkdir $masterdir file mkdir $clientdir set m_logtype [lindex $logset 0] set c_logtype [lindex $logset 1] # In-memory logs require a large log buffer, and cannot # be used with -txn nosync. set m_logargs [adjust_logargs $m_logtype] set c_logargs [adjust_logargs $c_logtype] set m_txnargs [adjust_txnargs $m_logtype] set c_txnargs [adjust_txnargs $c_logtype] # Open a master. repladd 1 set ma_cmd "berkdb_env_noerr -create $verbargs \ -log_max 1000000 $m_txnargs $m_logargs $repmemargs \ -home $masterdir -rep_master -errpfx MASTER \ -rep_transport \[list 1 replsend\]" set masterenv [eval $ma_cmd $recargs] # Open a client repladd 2 set cl_cmd "berkdb_env_noerr -create -home $clientdir $verbargs \ $c_txnargs $c_logargs -rep_client -errpfx CLIENT $repmemargs \ -rep_transport \[list 2 replsend\]" set clientenv [eval $cl_cmd $recargs] # Bring the client online. process_msgs "{$masterenv 1} {$clientenv 2}" # Open database in master, make lots of changes so checkpoint # will take a while, and propagate to client. puts "\tRep$tnum.a: Create and populate database." set dbname rep017.db set db [eval "berkdb_open_noerr -create $omethod -auto_commit \ -env $masterenv $largs $dbname"] for { set i 1 } { $i <= $niter } { incr i } { set t [$masterenv txn] error_check_good db_put \ [eval $db put -txn $t $i [chop_data $method data$i]] 0 error_check_good txn_commit [$t commit] 0 } process_msgs "{$masterenv 1} {$clientenv 2}" 1 # Get the master's last LSN before the checkpoint set pre_ckp_offset \ [stat_field $masterenv log_stat "Current log file offset"] puts "\tRep$tnum.b: Checkpoint on master." error_check_good checkpoint [$masterenv txn_checkpoint] 0 # Now get ckp LSN set ckp_lsn [stat_field $masterenv txn_stat "LSN of last checkpoint"] set ckp_offset [lindex $ckp_lsn 1] # Fork child process on client. It should process whatever # it finds in the message queue -- just the checkpoint record, # for now. It's run in the background so the parent can # test for whether we're checkpointing at the same time. # puts "\tRep$tnum.c: Fork child process on client." set pid [exec $tclsh_path $test_path/wrap.tcl \ rep017script.tcl $testdir/repscript.log \ $masterdir $clientdir $rep_verbose $verbose_type &] # We need to wait until we know that the client is processing a # checkpoint. The checkpoint will consist of some DBREG records # followed by the actual checkpoint. So, if we've gotten records # later than the last LSN when the master took the checkpoint, we've # begin the checkpoint. By test design, we should not finish the # checkpoint until this process has at least had a chance to run. # # In order to do this, we have handles open on the message # queue from both this process and its child. This is not # normally legal behavior for an application using Berkeley DB, # but this test depends on the parent process doing things while # the child is pausing in the middle of the checkpoint. We are # very careful to control which process is handling which # messages. puts "\tRep$tnum.d: Test whether client is in checkpoint." while { 1 } { set client_off \ [stat_field $clientenv log_stat "Current log file offset"] if { $client_off > $pre_ckp_offset } { if { $client_off > $ckp_offset } { # We already completed the checkpoint and # never got out of here. That's a bug in # in the test. error_check_good checkpoint_test \ not_in_checkpoint should_be_in_checkpoint } else { break; } } else { # Not yet up to checkpoint tclsleep 1 } } # Main client processes checkpoint 2nd time and should get NOTPERM. puts "\tRep$tnum.e: Commit and checkpoint return NOTPERM from client" incr niter set t [$masterenv txn] error_check_good db_put [eval $db put \ -txn $t $niter [chop_data $method data$niter]] 0 error_check_good txn_commit [$t commit] 0 error_check_good checkpoint [$masterenv txn_checkpoint] 0 set ckp2_lsn [stat_field $masterenv txn_stat "LSN of last checkpoint"] process_msgs "{$clientenv 2}" 1 # Check that the checkpoint record got a NOTPERM # Find the ckp LSN of the Master and then look for the response # from that message in the client set ckp_result "" foreach i $perm_response_list { # Everything in the list should be NOTPERM if { [llength $i] == 0 } { # Check for sentinel at beginning of list continue; } set ckp_result [lindex $i 0] error_check_good NOTPERM [is_substr $ckp_result NOTPERM] 1 if { [lindex $i 1] == $ckp2_lsn } { break } } error_check_bad perm_response $ckp_result "" puts "\tRep$tnum.f: Waiting for child ..." # Watch until the checkpoint is done. watch_procs $pid 5 # Verify that the checkpoint is now complete on the client and # that all later messages have been applied. process_msgs "{$clientenv 2}" 1 set client_ckp [stat_field $clientenv txn_stat "LSN of last checkpoint"] error_check_good matching_ckps $client_ckp $ckp2_lsn set m_end [stat_field $masterenv log_stat "Current log file offset"] set c_end [stat_field $clientenv log_stat "Current log file offset"] error_check_good matching_lsn $c_end $m_end # Finally, now that checkpoints are complete; perform another # perm operation and make sure that it returns ISPERM. puts "\tRep$tnum.g: No pending ckp; check for ISPERM" incr niter set t [$masterenv txn] error_check_good db_put [eval $db put \ -txn $t $niter [chop_data $method data$niter]] 0 error_check_good txn_commit [$t commit] 0 error_check_good checkpoint [$masterenv txn_checkpoint] 0 set ckp3_lsn [stat_field $masterenv txn_stat "LSN of last checkpoint"] process_msgs "{$clientenv 2}" 1 # Check that the checkpoint and commit records got a ISPERM # Find the ckp LSN of the Master and then look for the response # from that message in the client set ckp_result "" foreach i $perm_response_list { if { [llength $i] == 0 } { # Check for sentinel at beginning of list continue; } # Everything in the list should be ISPERM set ckp_result [lindex $i 0] error_check_good ISPERM [is_substr $ckp_result ISPERM] 1 if { [lindex $i 1] == $ckp3_lsn } { break } } error_check_bad perm_response $ckp_result "" # Clean up. error_check_good db_close [$db close] 0 error_check_good masterenv_close [$masterenv close] 0 error_check_good clientenv_close [$clientenv close] 0 replclose $testdir/MSGQUEUEDIR }