?> Op 27 okt. 2022 om 04:25 heeft Timo Sirainen <timo at sirainen.com> het volgende geschreven: > > ?Director never worked especially well, and for most use cases it's just unnecessarily complex. I think usually it could be replaced with: > > * Database (sql/ldap/whatever) containing user -> backend table. > * Configure Dovecot proxy to use this database as passdb. > * For HA change dovemon to update the database if backend is down to move users elsewhere > * When backend comes up, move users into it. Set delay_until extra field for user in passdb to 5 seconds into future and kick the user in its old backend (e.g. via doveadm HTTP API). > > All this can be done with existing Dovecot. Should be much easier to build a project doing this than forking director.This is my train of thought as well. I believe the following would suffice for most setups. A database with: - Current vhost count per backend server. Alternatively, count the temporary user mappings. - Backend servers. - Temporary user mappings between user - backend server. This database is accessible by all Dovecot proxies in case there?s multiple. Steps when receiving a login: - Check if a temporary user mapping exists. - If so, proxy to the backend server in the temporary mapping. (To do: clean up mappings.) - If not, pick the backend server with the lowest vhost count, create a temporary mapping, then increase the vhost count of the chosen backend server. A monitoring service up/downs backend servers. E.g. by checking the port that we proxy to for each backend server. When a backend server is set to down, kick the user to force a reconnection. (Is that how Director ?moves? users?)
On 2022-10-27 08:31, William Edwards wrote:> ? >> Op 27 okt. 2022 om 04:25 heeft Timo Sirainen <timo at sirainen.com> het >> volgende geschreven: >> >> ?Director never worked especially well, and for most use cases it's >> just unnecessarily complex. I think usually it could be replaced with: >> >> * Database (sql/ldap/whatever) containing user -> backend table. >> * Configure Dovecot proxy to use this database as passdb. >> * For HA change dovemon to update the database if backend is down to >> move users elsewhere >> * When backend comes up, move users into it. Set delay_until extra >> field for user in passdb to 5 seconds into future and kick the user in >> its old backend (e.g. via doveadm HTTP API). >> >> All this can be done with existing Dovecot. Should be much easier to >> build a project doing this than forking director. > > This is my train of thought as well. I believe the following would > suffice for most setups. > > A database with: > > - Current vhost count per backend server. Alternatively, count the > temporary user mappings. > - Backend servers. > - Temporary user mappings between user - backend server. > > This database is accessible by all Dovecot proxies in case there?s > multiple. > > Steps when receiving a login: > > - Check if a temporary user mapping exists. > - If so, proxy to the backend server in the temporary mapping. (To do: > clean up mappings.) > - If not, pick the backend server with the lowest vhost count, create a > temporary mapping, then increase the vhost count of the chosen backend > server. > > A monitoring service up/downs backend servers. E.g. by checking the > port that we proxy to for each backend server. When a backend server is > set to down, kick the user to force a reconnection. (Is that how > Director ?moves? users?)Here is my alternative input as well using database cluster/file. Create connection mappings table in database cluster where each row must be containing user id, backend id and frontend id and agent hash, alternatively mappings file containing such info and synced across all servers. Incorporate multiple simultaneous mappings using agent hash which can be useful e.g. in the event of using client apps from several devices, in the IMAP proxy perhaps update the first row agent hash which doesnt have hash and matching frontend and user id in post login requests. Create service in each backend, monitoring login and logout entries, and whenever there is one, add the relevant user and frontend row in mappings table/file. In the event of remove just mark one matching entry with exclusion to unknown agent hash as soft removed. In load balancing solution, for SMTP/IMAP connections, use perhaps a lua script, to check mappings in database or file, and find which backend user was logged to, and alongside generate user agent hash perhaps using base64 encoding to locate exact client connection backend row in mappings where several entries might be present, and proxy the incoming request to it, uncheck soft removed if same backend using same user agent hash, if there is no mappings, use the normal load balancing method which in post login requests its mappings will be automatically created. Zakaria.
On 2022-10-27 08:31, William Edwards wrote:> ? >> Op 27 okt. 2022 om 04:25 heeft Timo Sirainen <timo at sirainen.com> het >> volgende geschreven: >> >> ?Director never worked especially well, and for most use cases it's >> just unnecessarily complex. I think usually it could be replaced with: >> >> * Database (sql/ldap/whatever) containing user -> backend table. >> * Configure Dovecot proxy to use this database as passdb. >> * For HA change dovemon to update the database if backend is down to >> move users elsewhere >> * When backend comes up, move users into it. Set delay_until extra >> field for user in passdb to 5 seconds into future and kick the user in >> its old backend (e.g. via doveadm HTTP API). >> >> All this can be done with existing Dovecot. Should be much easier to >> build a project doing this than forking director. > > This is my train of thought as well. I believe the following would > suffice for most setups. > > A database with: > > - Current vhost count per backend server. Alternatively, count the > temporary user mappings. > - Backend servers. > - Temporary user mappings between user - backend server. > > This database is accessible by all Dovecot proxies in case there?s > multiple. > > Steps when receiving a login: > > - Check if a temporary user mapping exists. > - If so, proxy to the backend server in the temporary mapping. (To do: > clean up mappings.) > - If not, pick the backend server with the lowest vhost count, create a > temporary mapping, then increase the vhost count of the chosen backend > server. > > A monitoring service up/downs backend servers. E.g. by checking the > port that we proxy to for each backend server. When a backend server is > set to down, kick the user to force a reconnection. (Is that how > Director ?moves? users?)Here is my alternative input as well using database cluster/file. Create connection mappings table in database cluster where each row must be containing user id, backend id and frontend id and agent hash, alternatively mappings file containing such info and synced across all servers. Incorporate multiple simultaneous mappings using agent hash which can be useful e.g. in the event of using client apps from several devices, in the IMAP proxy perhaps update the first row agent hash which doesnt have hash and matching frontend and user id in post login requests. Create service in each backend, monitoring login and logout entries, and whenever there is one, add the relevant user and frontend row in mappings table/file. In the event of remove just mark one matching entry with exclusion to unknown agent hash as soft removed. In load balancing solution, for SMTP/IMAP connections, use perhaps a lua script, to check mappings in database or file, and find which backend user was logged to, and alongside generate user agent hash perhaps using base64 encoding to locate exact client connection backend row in mappings where several entries might be present, and proxy the incoming request to it, uncheck soft removed if same backend using same user agent hash, if there is no mappings, use the normal load balancing method which in post login requests its mappings will be automatically created. Zakaria.