| home | help | back | first | fref | pref | prev | next | nref | lref | last | post |
Date: Tue, 4 Feb 92 09:38:10 GMT From: Stig Ostholm <ostholm@ce.chalmers.se> To: info-afs@transarc.com A browse through the afs kernel code to pinpoint the problem with calls to a non-existing server/cell has given the following results: In the function `afs_NewCell': 1) A cell can not be deleted from the cache manager structures, only changed. The only way to modify a cell is through the AFSOP_ADDCELL call wich require at least one database server. 2) "Old" database servers are still present in the cache manager structures after an AFSOP_ADDCELL call is made for an existing cell with a list of new database server. The array `cellHost' of pointers to db servers is simply cleared and the new db servers are inserted. An old db server is still present in the `afs_servers' structure and points back to its former cell. If these observations are correct, could that be cause for the problem with clients trying to contact non-existing (removed) servers? Something like an afs call AFSOP_DELCELL that deletes all information about a cell from the cache manager should be added. All information about old database servers and all volume/server information for a deleted cell should be removed from the cache manager. We would like to know how non-DCE system are going to handle the information about cells/db-servers in the future. The current solution with a static text file containg all information has the same inconsistency problems as the host file before DNS. _ Stig Ostholm & Hakan Torbjar Chalmers University of Technology
| home | help | back | first | fref | pref | prev | next | nref | lref | last | post |