SDP-721
Open
Document steps for installing SDP on replica with NFS-shared /p4/common.
This includes:
* Getting /hxdepots/sdp correct.
* Getting /hxdepots/p4/common correct.
* Getting /p4/N (N=instance name) correct, containing
- /p4/N/bin
- /p4/N symlinks for root, offline_db, tmp, logs, depots, checkpooints, etc.
* Create /etc/systemd/system *.service files.
A bit of a preview of what I intend to document:
There are also gotchas to be aware of. It's what one would expect with NFS-sharing in general. For example, if you're on the backup server and you edit a config file in an NFS-shared directory -- you just edited the same file used by the primary server! Even experienced admins can easily forget the implications of NFS-sharing, as they are a bit counter-intuitive if your experience is mainly with fully duplicated environments.
NFS sharing is suitable for High Availability (HA) solutions only, not Disaster Recovery (DR) solutions, as DR solutions imply distances over which NFS sharing is not practical (and usually not possible).
Docs will also discuss some of the pros/cons.
Pros:
* You typically get other benefits from NFS hardware (e.g. snapshot capability).
* HA failover is simpler because you have zero chance of commits that didn't replicate.
* You don't need an extra full copy of whatever goes on /hxdepots (versioned files, checkpoints).
Cons:
* With NFS, there is now a single point of failure you didn't have before -- the NIC card on the NFS device. That risk is usually mitigated to some extent, as NFS devices are generally deemed "failure tolerant" because the vendors who produce them (e.g. NetApp) invest a lot to make them so.
* If you do suffer a failure of the NFS or the data on it that cannot be recovered easily, you must do a DR failover rather than an HR failover. (For many sites, HA failover is more likely to work than DR failover because DR involves new network paths from users to servers, servers to integrated systems, and other complexities that should be accounted for in a comprehensive failover plan).
* NFS environments have a slightly higher incidence of "operator error" until admins get comfortable dealing with the aforementioned counter-intuitive implications of NFS sharing.
This includes:
* Getting /hxdepots/sdp correct.
* Getting /hxdepots/p4/common correct.
* Getting /p4/N (N=instance name) correct, containing
- /p4/N/bin
- /p4/N symlinks for root, offline_db, tmp, logs, depots, checkpooints, etc.
* Create /etc/systemd/system *.service files.
A bit of a preview of what I intend to document:
There are also gotchas to be aware of. It's what one would expect with NFS-sharing in general. For example, if you're on the backup server and you edit a config file in an NFS-shared directory -- you just edited the same file used by the primary server! Even experienced admins can easily forget the implications of NFS-sharing, as they are a bit counter-intuitive if your experience is mainly with fully duplicated environments.
NFS sharing is suitable for High Availability (HA) solutions only, not Disaster Recovery (DR) solutions, as DR solutions imply distances over which NFS sharing is not practical (and usually not possible).
Docs will also discuss some of the pros/cons.
Pros:
* You typically get other benefits from NFS hardware (e.g. snapshot capability).
* HA failover is simpler because you have zero chance of commits that didn't replicate.
* You don't need an extra full copy of whatever goes on /hxdepots (versioned files, checkpoints).
Cons:
* With NFS, there is now a single point of failure you didn't have before -- the NIC card on the NFS device. That risk is usually mitigated to some extent, as NFS devices are generally deemed "failure tolerant" because the vendors who produce them (e.g. NetApp) invest a lot to make them so.
* If you do suffer a failure of the NFS or the data on it that cannot be recovered easily, you must do a DR failover rather than an HR failover. (For many sites, HA failover is more likely to work than DR failover because DR involves new network paths from users to servers, servers to integrated systems, and other complexities that should be accounted for in a comprehensive failover plan).
* NFS environments have a slightly higher incidence of "operator error" until admins get comfortable dealing with the aforementioned counter-intuitive implications of NFS sharing.
- Status
- Open
- Project
- perforce-software-sdp
- Severity
- C
- Reported By
- lee_marzke
- Reported Date
- Modified By
- C. Thomas Tyler
- Modified Date
- Owned By
- tom_tyler
- Component
- doc
- Type
- Feature