(ITS#7062) writes from lost connections not increasing contextCSN

[ebackes@symas.com]

Full_Name: Emily Backes
Version: 2.4.26
OS: any
URL: 
Submission from: (NULL) (75.244.124.114)


If a client connects and writes, e.g. ADDs entries, then disconnects
before seeing the result, the writes are stored in the database but do
not increment contextCSN.

This can be seen without replicas being involved, but obviously still
needs syncprov.

$ ldapadd  -x -D dc=example,dc=com -w secret -H ldap://localhost:1389/ -f
people.ldif
adding new entry "cn=1,ou=people,dc=example,dc=com"

adding new entry "cn=2,ou=people,dc=example,dc=com"
...
adding new entry "cn=22,ou=people,dc=example,dc=com"
^C

$ ldapsearch -x -D dc=example,dc=com -w secret -H ldap://localhost:1389/ -b
dc=example,dc=com -LLL '(cn=22)' entryCSN
dn: cn=22,ou=people,dc=example,dc=com
entryCSN: 20111011235653.469937Z#000000#001#000000

$ ldapsearch -x -D dc=example,dc=com -w secret -H ldap://localhost:1389/ -b
dc=example,dc=com -LLL -s base contextCSN
dn: dc=example,dc=com
contextCSN: 20111011235653.380889Z#000000#001#000000

stats+sync log:

conn=1000 op=21 ADD dn="cn=21,ou=people,dc=example,dc=com"
slap_queue_csn: queing 0x105e00100 20111011235653.380889Z#000000#001#000000
conn=1000 op=21 RESULT tag=105 err=0 text=
slap_graduate_commit_csn: removing 0x10178e0a0
20111011235653.380889Z#000000#001#000000
conn=1000 op=22 ADD dn="cn=22,ou=people,dc=example,dc=com"
slap_queue_csn: queing 0x105e00100 20111011235653.469937Z#000000#001#000000
slap_graduate_commit_csn: removing 0x10178e520
20111011235653.469937Z#000000#001#000000
conn=1000 fd=10 closed (connection lost)

The change isn't sent out to connected refreshAndPersist replicas
either, and forcing them to reconnect doesn't help, as their
contextCSNs still match across the board.  This can cause persistent
desynchronization of servers.