Secondary Storage
Secondary or cold storage is intended for usages where the primary or hot storage is low-latency and fast, such as NVME, and the secondary is high-latency but very large, such as a SAN spanning many spinning disks. Only segment files get moved to the secondary storage. Older files get moved before younger ones.
Files Moved based on Available Disk Space
When enabled, LogScale will move segment files to secondary storage
once the primary disk reaches the usage threshold set using
PRIMARY_STORAGE_PERCENTAGE
. LogScale does not check
what is using the space; it bases the decision on what the OS
responds for "disk space used" and "disk space total" for the mount
point that the primary data directory is on.
When the threshold is exceeded, LogScale will copy files totaling the excess number of bytes to the secondary storage, and then delete the segment files from the primary data directory. The files are selected based on the latest event timestamp in them, to keep most recent events on the primary disk. This is done to get the best possible query performance from the assumed faster primary drive, since LogScale is normally used for querying the latest data.
The extra storage gained is thus almost the available space on of the secondary data directory, as only a single segment file is ever present on both volumes at once.
The secondary directory needs to be private to the LogScale node, just like the primary directory does.
Note
Never share data directories across LogScale nodes.
As an example, suppose you have a server with 1 TB NVME being used for system files, Kafka data, and LogScale data. Adding a 2 TB SAN connection (or 2x2 TB local spinning disks in a mirror) and then designating that as secondary storage directory allows LogScale to store up to 2.8 TB, while still querying the latest ~800 GB from the NVME, and also keeping all segment files still being constructed on the NVME. When searching beyond what the NVME holds, LogScale will read from the slower disks.
Configuring
LogScale needs to be told where to store the secondary copies; that is, the location of the filesystem on the slower drive. When using Docker, make sure to mount the secondary directory into the container as well.
# SECONDARY_DATA_DIRECTORY enables the feature
# and sets where to store the files.
SECONDARY_DATA_DIRECTORY=/secondaryMountPoint/humio-data2
# PRIMARY_STORAGE_PERCENTAGE options decide the amount of data (LogScale
# and otherwise) that the drive holding the data directory must at least hold
# before LogScale decides to move any segments files to the secondary location.
# If set to zero, LogScale will move files to secondary as soon as possible,
# which is when they become immutable completed segment files.
# (Default 80)
PRIMARY_STORAGE_PERCENTAGE=80
Adding a New Primary and Making the Current Primary Disk Secondary
Say you have a slow disk as your only (and thus primary) disk for LogScale. You add a new faster disk to the server, and want to use that disk as the primary, while leaving the bulk of the data on the old slow disk.
While this is possible, there is a bit of work involved, as only completed segment files can reside in the secondary storage. All other support files, and segment files in progress, need to reside on the primary disk. LogScale must be shut down while this operation takes place.
Basically only files matching humiodata.*
(and bloom5*
) can stay on
the secondary storage; everything else must be on the primary. The
tricky bit is moving the soft links
humiodata.current
along with the file they
point to.
You will need to move some specific files from the "new secondary"
onto the "new primary" while the system is shut down for that to
work, as some files must be on the primary. Here are their names,
as they are below /humio-data
. The directory
structure must be preserved.
Files matching
dataspace_*/datasource_*/humiodata.current
For all the above
humiodata.current
soft links, the file it points to as well.uploadedfiles
directorycluster_membership.uuid
fileglobal-data-snapshot.json
file
If the above files are moved from the secondary to the primary, you may leave the remaining segment files, and start out with almost all data being on secondary. Or, if you want, move selected parts of the completed segment files from secondary to primary as well, to get improved performance from the new disk on searches that hit those. One could move all segments that are less than seven days old if that matches the search typical search range for the system.
LogScale will not move files from secondary back to primary. Once the primary is full, LogScale will start migrating segment files from primary to secondary.
Migrating Storage Directories
When expanding the capacity for primary or secondary storage, and online expansion is not an option, you can move the existing data stored on the partitions and minimize downtime using rsync. Rsync allows you to sync only the new data between directories.
Assuming you're moving secondary data storage from
/var/lib/humio-secondary
to
/var/lib/humio-secondary-new
you can do an
initial rsync while LogScale is running. Make
sure your new mount has its owner and user set appropriately,
generally humio:humio
.
rsync -acv /var/lib/humio-secondary/ /var/lib/humio-new-secondary/
This can be run multiple times.
When you are ready to complete the move, start by stopping LogScale:
systemctl stop humio
This command may differ depending on your LogScale deployment. To move the data written since your last rsync, the delete option will remove no longer needed files from the destination. Double check your source and destination directories.
rsync -acv --delete /var/lib/humio-secondary/ /var/lib/humio-new-secondary/
There are two options for having LogScale use the new partition, you can either update the LogScale configuration to use the new mount or you can remount the new partition in the place of the old partitions mount point.
To update the LogScale configuration appropriately, change either
the value of the DIRECTORY
or
SECONDARY_DATA_DIRECTORY
option in
/etc/humio/server_XX.conf
file to point to the
new mount. Then restart LogScale:
systemctl start humio
To unmount and mount the new partition in place of LogScale, make
sure first that the appropriate changes are made for your OS, such
as updating the /etc/fstab
, and mount the new
partition in the location of the old one. Once that's done, start
LogScale.