# See the file LICENSE for redistribution information. # # Copyright (c) 2010, 2011 Oracle and/or its affiliates. All rights reserved. # # TEST rep091 # TEST Read-your-writes consistency. # TEST Write transactions at the master, and then call the txn_applied() # TEST method to see whether the client has received and applied them yet. # proc rep091 { method { niter 20 } { tnum "091" } args } { source ./include.tcl global databases_in_memory global repfiles_in_memory # Skip for all methods except btree. if { $checking_valid_methods } { set test_methods { btree } return $test_methods } if { [is_btree $method] == 0 } { puts "Rep091: Skipping for method $method." return } # Set up for on-disk or in-memory databases and make dbname match # the one assigned in rep_test. if { $databases_in_memory == 1 } { set dbname { "" "test.db" } set msg "using named in-memory databases" } else { set dbname "test.db" set msg "using on-disk databases" } # For this test it's important to test both ways, so for now run both # under our control. Later, when the containing test infrastructure for # doing this automatically is more fully developed, we can remove this # loop and just let the infrastructure handle it. # set orig $repfiles_in_memory foreach repfiles_in_memory {0 1} { if { $repfiles_in_memory } { set msg2 "and in-memory replication files" } else { set msg2 "and on-disk replication files" } puts "Rep$tnum ($method $args): Test of\ read-your-writes consistency $msg $msg2." rep091a_sub $method $niter $tnum no $dbname $args rep091a_sub $method $niter $tnum yes $dbname $args rep091b_sub $method $niter $tnum $dbname $args rep091c_sub $method $niter $tnum no $dbname $args rep091c_sub $method $niter $tnum yes $dbname $args rep091d_sub $method $niter $tnum no $dbname $args rep091d_sub $method $niter $tnum yes $dbname $args rep091e_sub $method $niter $tnum $dbname $args if { $repfiles_in_memory } { rep091f_sub $method $niter $tnum $dbname $args } } # Restore original setting, in case any other tests are going to be run # after this one. # set repfiles_in_memory $orig } proc rep091a_sub { method niter tnum future_gen dbname largs } { global rep_verbose global testdir global verbose_type global repfiles_in_memory puts -nonewline "Rep$tnum: read-your-writes consistency, basic test" if { $future_gen } { puts ", with extra gen cycle." } else { puts "." } 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 replsetup $testdir/MSGQUEUEDIR set masterdir $testdir/MASTERDIR set clientdir $testdir/CLIENTDIR file mkdir $masterdir file mkdir $clientdir puts "\tRep$tnum.a: Create master and client." repladd 1 set ma_envcmd "berkdb_env_noerr -create -txn -errpfx MASTER \ $repmemargs \ $verbargs -home $masterdir -rep_transport \[list 1 replsend\]" set masterenv [eval $ma_envcmd -rep_master] repladd 2 set cl_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT \ $repmemargs \ $verbargs -home $clientdir -rep_transport \[list 2 replsend\]" set clientenv [eval $cl_envcmd -rep_client] set envlist "{$masterenv 1} {$clientenv 2}" process_msgs $envlist if { $future_gen } { # Cycle the gen an extra time, so that the client sees not only # a future transaction, but a future gen. Restart the master as # a client first, and communicate with the surviving other # client, to make sure the master realizes the correct gen. We # can get away without doing this in the normal case, but when # repfiles are in memory the master otherwise starts over from # gen 0, leading to quite a mess. # $masterenv close set masterenv [eval $ma_envcmd -rep_client -recover] set envlist "{$masterenv 1} {$clientenv 2}" process_msgs $envlist $masterenv rep_start -master } set start 0 eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter puts "\tRep$tnum.b: Write txn and get its commit token." set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ -create $omethod $dbargs $dbname ] set txn [$masterenv txn -token] $db put -txn $txn "key1" "data1" set token [$txn commit] # binary scan $token IIIII r1 r2 r3 r4 r5 # puts "the token value is $r1 $r2 $r3 $r4 $r5" # For the most part, we're interested in checking txn_applied at # clients. But it's also supposed to work in a degenerate, simple # manner at the master. # error_check_good applied_at_master [$masterenv txn_applied $token] 0 # While we're at it, verify that if we don't ask for the commit token, # the result of [$env commit] is just a plain "0". (This is more of a # sanity check on the Tcl API, to help us believe that the rest of this # test is valid.) # set txn [$masterenv txn] $db put -txn $txn "other key" "any data" set result [$txn commit] error_check_good commit_no_token $result 0 $db close puts "\tRep$tnum.c: Check the applied status of the transaction at the client." # Before processing messages, the client will not have heard of the # transaction yet. But afterwards it should. Check it once with no # timeout (default 0), and again with a timeout specified to do a simple # test of waiting. (Multi-thread waiting tests will be done # separately.) # set result [$clientenv txn_applied $token] error_check_good not_applied [is_substr $result DB_TIMEOUT] 1 set start [clock seconds] set result [$clientenv txn_applied -timeout 10000000 $token] set duration [expr [clock seconds] - $start] error_check_good not_yet_applied [expr $duration >= 10] 1 process_msgs $envlist error_check_good txn_applied [$clientenv txn_applied $token] 0 # "Empty" transactions are not really interesting either, but again # they're supposed to be allowed: # set txn [$masterenv txn -token] set token [$txn commit] set result [$masterenv txn_applied $token] error_check_good empty_txn [is_substr $result DB_KEYEMPTY] 1 set result [$clientenv txn_applied $token] error_check_good empty_txn2 [is_substr $result DB_KEYEMPTY] 1 # Check a few simple invalid cases. # error_check_bad client_token [catch {$clientenv txn -token} result] 0 error_check_good cl_tok_msg [is_substr $result "invalid arg"] 1 set txn [$masterenv txn] error_check_bad child_token \ [catch {$masterenv txn -token -parent $txn} result] 0 error_check_good parent_tok_msg [is_substr $result "invalid arg"] 1 $txn abort $clientenv close $masterenv close replclose $testdir/MSGQUEUEDIR } # Verify that an LSN history database appears correctly at a client even if the # client is created via a hot backup (instead of internal init). Sites "A" and # "B" alternate the master role, in order to facilitate accumulating a history # of gen changes (especially in the in-mem case, where otherwise our LSN history # database would be lost each time we restarted the master). The "CLIENT2" site # will be the one to be initialized via hot backup. # proc rep091b_sub { method niter tnum dbname largs } { global rep_verbose global testdir global verbose_type global util_path global repfiles_in_memory puts "Rep$tnum: read-your-writes consistency, hot backup." 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 replsetup $testdir/MSGQUEUEDIR set dira $testdir/DIRA set dirb $testdir/DIRB set clientdir2 $testdir/CLIENTDIR2 file mkdir $dira file mkdir $dirb file mkdir $clientdir2 repladd 1 set envcmd_a "berkdb_env_noerr -create -txn -errpfx SITE_A \ $repmemargs \ $verbargs -home $dira -rep_transport \[list 1 replsend\]" set masterenv [eval $envcmd_a -rep_master] repladd 2 set envcmd_b "berkdb_env_noerr -create -txn -errpfx SITE_B \ $repmemargs \ $verbargs -home $dirb -rep_transport \[list 2 replsend\]" set clientenv [eval $envcmd_b -rep_client] set envlist "{$masterenv 1} {$clientenv 2}" process_msgs $envlist set start 0 # Swap master and client role, some arbitrary number of times, in order # to change the generation number, before writing our transaction of # interest. # set count 2 for { set i 0 } { $i < $count } { incr i } { eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist set tmp $masterenv set masterenv $clientenv set clientenv $tmp $clientenv rep_start -client $masterenv rep_start -master process_msgs $envlist } set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$masterenv txn -token] $db put -txn $txn "some key" "some data" set token [$txn commit] eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist # Cycle the master gen a few more times, in order to force the token gen # not to be at the top of the history. Again, the count is arbitrary. # $db close set count 3 for { set i 0 } { $i < $count } { incr i } { set tmp $masterenv set masterenv $clientenv set clientenv $tmp $clientenv rep_start -client $masterenv rep_start -master process_msgs $envlist eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist } # Note that the choice of which existing directory ("a" or "b") we use # as the source of the copy doesn't matter. # exec $util_path/db_hotbackup -h $dira -b $clientdir2 set count 2 for { set i 0 } { $i < $count } { incr i } { set tmp $masterenv set masterenv $clientenv set clientenv $tmp $clientenv rep_start -client $masterenv rep_start -master process_msgs $envlist eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist } # Create a second transaction, after a few more master gen changes, and # get its token too. # set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$masterenv txn -token] $db put -txn $txn "other key" "different data" set token2 [$txn commit] # Start the new client, by running "catastrophic" recovery on the # hot-backup files. # exec $util_path/db_recover -c -h $clientdir2 repladd 3 set cl2_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT2 \ $repmemargs \ $verbargs -home $clientdir2 -rep_transport \[list 3 replsend\]" set clientenv2 [eval $cl2_envcmd -rep_client] $clientenv2 rep_config {autoinit off} # Before syncing with the master, we should be able to confirm the first # transaction, but not the second one, because the hot backup should # include an (old) copy of the LSN history database. # if {$repfiles_in_memory} { set result [$clientenv2 txn_applied $token] error_check_good still_waiting [is_substr $result DB_TIMEOUT] 1 set result2 [$clientenv2 txn_applied $token2] error_check_good not_yet_applied [is_substr $result2 DB_TIMEOUT] 1 } else { error_check_good txn_applied [$clientenv2 txn_applied $token] 0 set result2 [$clientenv2 txn_applied $token2] error_check_good not_yet_applied [is_substr $result2 DB_NOTFOUND] 1 } # Sync with master, and this time token2 should be there. # lappend envlist [list $clientenv2 3] process_msgs $envlist error_check_good txn_applied2 [$clientenv2 txn_applied $token2] 0 $db close $clientenv2 close $clientenv close $masterenv close replclose $testdir/MSGQUEUEDIR } # Test detection of rollbacks. # proc rep091c_sub { method niter tnum future_gen dbname largs } { global rep_verbose global testdir global verbose_type global repfiles_in_memory puts "Rep$tnum: read-your-writes consistency, rollbacks." 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 replsetup $testdir/MSGQUEUEDIR set masterdir $testdir/MASTERDIR set clientdir $testdir/CLIENTDIR set clientdir2 $testdir/CLIENTDIR2 file mkdir $masterdir file mkdir $clientdir file mkdir $clientdir2 puts "\tRep$tnum.a: Create a group of three." repladd 1 set ma_envcmd "berkdb_env_noerr -create -txn -errpfx MASTER \ $repmemargs \ $verbargs -home $masterdir -rep_transport \[list 1 replsend\]" set masterenv [eval $ma_envcmd -rep_master] repladd 2 set cl_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT \ $repmemargs \ $verbargs -home $clientdir -rep_transport \[list 2 replsend\]" set clientenv [eval $cl_envcmd -rep_client] repladd 3 set cl2_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT2 \ $repmemargs \ $verbargs -home $clientdir2 -rep_transport \[list 3 replsend\]" set clientenv2 [eval $cl2_envcmd -rep_client] set envlist "{$masterenv 1} {$clientenv 2} {$clientenv2 3}" process_msgs $envlist set start 0 eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist # If desired, test the case where the token has a "future" gen, by # bumping of the gen a few times after cutting off the disconnected # client. # if { $future_gen } { for { set count 3 } { $count > 0 } { incr count -1 } { $masterenv close set masterenv [eval $ma_envcmd -rep_master -recover] eval rep_test $method $masterenv NULL \ $niter $start $start 0 $largs incr start $niter } set envlist "{$masterenv 1} {$clientenv 2} {$clientenv2 3}" } # Write some more transactions, taking a token for one of them, but # prevent one of the clients from seeing any of them. # puts "\tRep$tnum.b: Write transactions, and get token." eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$masterenv txn -token] $db put -txn $txn "some key" "some data" set token [$txn commit] $db close eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter replclear 2 process_msgs $envlist puts "\tRep$tnum.c: Switch master, forcing a rollback." $masterenv close # $clientenv rep_start -master $clientenv rep_start -client; # set envlist "{$clientenv 2} {$clientenv2 3}"; # process_msgs $envlist; # $clientenv rep_start -master; # eval rep_test $method $clientenv NULL $niter $start $start 0 $largs incr start $niter replclear 1 # set envlist "{$clientenv 2} {$clientenv2 3}" process_msgs $envlist puts "\tRep$tnum.d: Check txn_applied at client and at (new) master." set result [$clientenv2 txn_applied $token] error_check_good rolled_back [is_substr $result DB_NOTFOUND] 1 set result [$clientenv txn_applied $token] error_check_good rolled_back2 [is_substr $result DB_NOTFOUND] 1 $clientenv close $clientenv2 close replclose $testdir/MSGQUEUEDIR } # Test envid check. Simulate a network partition: two masters proceed at the # same gen, in the same LSN range. The envid would be the only way we would # know that a transaction from the disconnected master is not correctly applied # at a client in the other partition. # proc rep091d_sub { method niter tnum extra_gen dbname largs } { global rep_verbose global testdir global verbose_type global repfiles_in_memory puts -nonewline "Rep$tnum: read-your-writes consistency, partition" if { $extra_gen } { puts ", with extra gen." } else { puts "." } 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 replsetup $testdir/MSGQUEUEDIR set masterdir $testdir/MASTERDIR set clientdir $testdir/CLIENTDIR set clientdir2 $testdir/CLIENTDIR2 file mkdir $masterdir file mkdir $clientdir file mkdir $clientdir2 puts "\tRep$tnum.a: Create a group of three." repladd 1 set ma_envcmd "berkdb_env_noerr -create -txn -errpfx MASTER \ $repmemargs \ $verbargs -home $masterdir -rep_transport \[list 1 replsend\]" set masterenv [eval $ma_envcmd -rep_master] repladd 2 set cl_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT \ $repmemargs \ $verbargs -home $clientdir -rep_transport \[list 2 replsend\]" set clientenv [eval $cl_envcmd -rep_client] repladd 3 set cl2_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT2 \ $repmemargs \ $verbargs -home $clientdir2 -rep_transport \[list 3 replsend\]" set clientenv2 [eval $cl2_envcmd -rep_client] set envlist "{$masterenv 1} {$clientenv 2} {$clientenv2 3}" process_msgs $envlist set start 0 eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist # Bounce master and client1. Then when they come back up, pretend # client1 is disconnected, and somehow decides to act as a master (an # application not using elections, obviously). Note that we must # replclear appropriately in both directions. # $masterenv close $clientenv close set masterenv [eval $ma_envcmd -rep_master -recover] replclear 2 set clientenv [eval $cl_envcmd -rep_master -recover] replclear 1 replclear 3 set envlist "{$masterenv 1} {$clientenv2 3}" puts "\tRep$tnum.b: Run identical series of txns at two masters." set orig_start $start eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$masterenv txn] $db put -txn $txn "some key" "some data" $txn commit $db close eval rep_test $method $masterenv NULL $niter $start $start 0 $largs replclear 2 process_msgs $envlist set start $orig_start eval rep_test $method $clientenv NULL $niter $start $start 0 $largs incr start $niter set db [eval berkdb_open_noerr -env $clientenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$clientenv txn -token] $db put -txn $txn "some key" "some data" set token [$txn commit] $db close eval rep_test $method $clientenv NULL $niter $start $start 0 $largs incr start $niter replclear 1 replclear 3 if { $extra_gen } { $masterenv close set masterenv [eval $ma_envcmd -rep_master -recover] set envlist "{$masterenv 1} {$clientenv2 3}" eval rep_test $method $masterenv NULL $niter $start $start 0 $largs replclear 2 process_msgs $envlist } puts "\tRep$tnum.c: Check txn_applied." set result [$clientenv2 txn_applied $token] error_check_good not_found [is_substr $result DB_NOTFOUND] 1 $clientenv close $clientenv2 close $masterenv close replclose $testdir/MSGQUEUEDIR } # Test the simplified, degenerate behavior of a non-replication environment: a # valid transaction is always considered applied, unless it disappears in a # backup/restore. # proc rep091e_sub { method niter tnum dbname largs } { global testdir global util_path set backupdir "$testdir/backup" file mkdir $backupdir puts "Rep$tnum: read-your-writes consistency, non-rep env." env_cleanup $testdir set dir $testdir set envcmd "berkdb_env_noerr -create -txn -home $dir" set env [eval $envcmd] set start 0 eval rep_test $method $env NULL $niter $start $start 0 $largs incr start $niter exec $util_path/db_hotbackup -h $testdir -b $backupdir set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $env -auto_commit \ $omethod $dbargs $dbname ] set txn [$env txn -token] $db put -txn $txn "some key" "some data" set token [$txn commit] $db close eval rep_test $method $env NULL $niter $start $start 0 $largs incr start $niter error_check_good applied [$env txn_applied $token] 0 $env close set envcmd "berkdb_env_noerr -create -txn -home $backupdir -recover_fatal" set env [eval $envcmd] set result [$env txn_applied $token] error_check_good restored [is_substr $result DB_NOTFOUND] 1 $env close } # Check proper behavior of txn_applied() at a client before in-mem DB's have # been materialized. This is only relevant when repfiles_in_memory, so skip it # in the default case. Sites "A" and "B" will take turns being master, and # CLIENT2 will be the one whose behavior is under test. # proc rep091f_sub { method niter tnum dbname largs } { global rep_verbose global testdir global verbose_type global util_path global repfiles_in_memory if { $repfiles_in_memory } { set repmemargs "-rep_inmem_files " puts "Rep$tnum: read-your-writes consistency, missing in-mem DB." } else { puts "Rep$tnum: skipping missing in-mem DB test." return } set verbargs "" if { $rep_verbose == 1 } { set verbargs " -verbose {$verbose_type on} " } env_cleanup $testdir replsetup $testdir/MSGQUEUEDIR set dira $testdir/DIRA set dirb $testdir/DIRB set clientdir2 $testdir/CLIENTDIR2 file mkdir $dira file mkdir $dirb file mkdir $clientdir2 repladd 1 set envcmd_a "berkdb_env_noerr -create -txn -errpfx SITE_A \ $repmemargs \ $verbargs -home $dira -rep_transport \[list 1 replsend\]" set masterenv [eval $envcmd_a -rep_master] repladd 2 set envcmd_b "berkdb_env_noerr -create -txn -errpfx SITE_B \ $repmemargs \ $verbargs -home $dirb -rep_transport \[list 2 replsend\]" set clientenv [eval $envcmd_b -rep_client] repladd 3 set cl2_envcmd "berkdb_env_noerr -create -txn -errpfx CLIENT2 \ $repmemargs \ $verbargs -home $clientdir2 -rep_transport \[list 3 replsend\]" set clientenv2 [eval $cl2_envcmd -rep_client] set envlist "{$masterenv 1} {$clientenv 2} {$clientenv2 3}" process_msgs $envlist set start 0 eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter # Write the transaction of interest in gen 1. # set omethod [convert_method $method] set dbargs [convert_args $method $largs] set db [eval berkdb_open_noerr -env $masterenv -auto_commit \ $omethod $dbargs $dbname ] set txn [$masterenv txn -token] $db put -txn $txn "some key" "some data" set token [$txn commit] $db close eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist set tmp $masterenv set masterenv $clientenv set clientenv $tmp $clientenv rep_start -client $masterenv rep_start -master process_msgs $envlist eval rep_test $method $masterenv NULL $niter $start $start 0 $largs incr start $niter process_msgs $envlist # Bounce the client. Even though it has the transaction of interest, # and had observed the change to gen 2, when we restart it it won't have # its LSN history database. # $clientenv2 close set clientenv2 [eval $cl2_envcmd -rep_client -recover] set result [$clientenv2 txn_applied $token] error_check_good not_yet [is_substr $result DB_TIMEOUT] 1 # Sync with master, and this time token should be confirmed. # set envlist [lreplace $envlist end end [list $clientenv2 3]] process_msgs $envlist error_check_good txn_applied [$clientenv2 txn_applied $token] 0 $clientenv2 close $clientenv close $masterenv close replclose $testdir/MSGQUEUEDIR }