Overview
From version 4.6, Artifactory offers a Sharding Binary Provider that lets you manage your binaries in a sharded filestore. A sharded filestore is one that is implemented on a number of physical mounts (M), which store binary objects with redundancy (R), where R <= M.
For example, the diagram below represents a sharded filestore where M=3 and R=2. In other words, the filestore consists of 3 physical mounts which store each binary in two copies.
Artifactory’s sharding binary provider presents several benefits:
Unmatched stability and reliability
Thanks to redundant storage of binaries, the system can withstand any mount going down as long as M >= R.
Unlimited scalability
If the underlying storage available approaches depletion, you only need to add another mount; a process that requiresno downtimeof the filestore. Once the mount is up and running, the system regenerates the filestore redundancy according to configuration parameters you control.
For a standalone Artifactory setup
A restart is required to apply the changes to yourbinarystore.xml settings.
For a High Availablity setup
Restarting each cluster separately to apply the changes to yourbinarystore.xmlsettings will result in no downtime.
Filestore performance optimization
Sharding Binary Provider offers several configuration parameters that allow you to optimize how binaries are read from or written to the filestore according to your specific system’s requirements.
Enterprise license required
Sharded filestore is available for Artifactory installations activated with an enterprise license.
Configuring a Sharding Binary Provider
A sharding binary provider is a binary provider as described inConfiguring the Filestore.Basic sharding configurationis used to configure a sharding binary provider for an instance of Artifactory Pro.
Basic Sharding Configuration
The following parameters are available for a basic sharding configuration:
lenientLimit |
Default: 1 (From version 5.4. Note that for filestores configured with a custom chain and not using the built-in templates the default value of the Theminimumnumber of successful writes thatmust be maintained for an upload to be successful. The next balance cycle (triggered with the GC mechanism) will eventually transfer the binary to enough nodes such that the redundancy commitment is preserved. For example, if The amount of currently active nodes must always be greater or equal than the configured lenientLimit. If set to 0, the redundancy value has to be kept. |
readBehavior |
This parameter dictates the strategy for reading binaries from the mounts that make up the sharded filestore. Possible values are: roundRobin (default):Binaries are read from each mount using a round robin strategy. |
writeBehavior |
This parameter dictates the strategy for writing binaries to the mounts that make up the sharded filestore. Possible values are: roundRobin (default):Binaries are written to each mount using a round robin strategy. freeSpace:Binaries are written to the mount with the greatest absolute volume of free space available. percentageFreeSpace:Binaries are written to the mount with the percentage of free space available. |
redundancy |
Default: r=1 The number of copies that should be stored for each binary in the filestore. Note that redundancy must be less than or equal to the number of mounts in your system for Artifactory to work with this configuration. |
lenientLimit |
Default: 1 (From version 5.4. Note that for filestores configured with a custom chain and not using the built-in templates the default value of the Theminimumnumber of successful writes thatmust be maintained for an upload to be successful. The next balance cycle (triggered with the GC mechanism) will eventually transfer the binary to enough nodes such that the redundancy commitment is preserved. For example, if The amount of currently active nodes must always be greater or equal than the configured lenientLimit. If set to 0, the redundancy value has to be kept. |
concurrentStreamWaitTimeout |
Default: 30,000 ms To support the specified redundancy, accumulates the write stream in a buffer, and uses “r” threads (according to the specified redundancy) to write to each of the redundant copies of the binary being written. A binary can only be considered written once all redundant threads have completed their write operation. Since all threads are competing for the write stream buffer, each one will complete the write operation at a different time. This parameter specifies the amount of time (ms) that any thread will wait for all the others to complete their write operation. If a write operation fails, you can try increasing the value of this parameter. |
concurrentStreamBufferKb |
Default: 32 Kb If a write operation fails, you can try increasing the value of this parameter. |
maxBalancingRunTime |
Default: 3,600,000 ms (1 hour)
To restore your system to full redundancy more quickly after a mount failure, you may increase the value of this parameter. If you find this causes an unacceptable degradation of overall system performance, you can consider decreasing the value of this parameter, but this means that the overall time taken for Artifactory to restore full redundancy will be longer.
|
freeSpaceSampleInterval |
Default: 3,600,000 ms (1 hour) To implement its write behavior, Artifactory needs to periodically query the mounts in the sharded filestore to check for free space. Since this check may be a resource intensive operation, you may use this parameter to control the time interval between free space checks.
If you anticipate a period of intensive upload of large volumes of binaries, you can consider decreasing the value of this parameter in order to reduce the transient imbalance between mounts in your system.
|
minSpareUploaderExecutor |
Default: 2 Artifactory maintains a pool of threads to execute writes to each redundant unit of storage. Depending on the intensity of write activity, eventually, some of the threads may become idle and are then candidates for being killed. However, Artifactory does need to maintain some threads alive for when write activities begin again. This parameter specifies the minimum number of threads that should be kept alive to supply redundant storage units. |
uploaderCleanupIdleTime |
Default: 120,000 ms (2 min) The maximum period of time threads may remain idle before becoming candidates for being killed. |
Example 1
The code snippet below is a sample configuration for the following setup:
- A cached sharding binary provider with three mounts and redundancy of 2.
- Each mount "X" writes to a directory called /filestoreX.
- The read strategy for the provider isroundRobin.
- The write strategy for the provider ispercentageFreeSpace.
<配置version = " 4 " > <链> <提供者id = " cache-fs" type="cache-fs">// Specify the read and write strategy and redundancy for the sharding binary provider //For each sub-provider (mount), specify the filestore location roundRobin percentageFreeSpace 2 filestore1 filestore2 filestore3
Example 2
The following code snippet shows the "double-shards" template which can be used as is for your binary store configuration.
<配置version = " 4 " > < =“double-shard链模板s" />shard-fs-1 shard-fs-2
The double-shards template uses a cached provider with two mounts and a redundancy of 1, i.e. only one copy of each artifact is stored.
1
To modify the parameters of the template, you can change the values of the elements in the template definition. For example, to increase redundancy of the configuration to 2, you only need to modify the
2
Cross-Zone Sharding Configuration
Sharding across multiple zones in an HA Artifactory cluster allows you to create zones or regions of sharded data to provide additional redundancy in case one of your zones becomes unavailable. You can determine the order in which the data is written between the zones and you can set the method for establishing the free space when writing to the mounts in the neighboring zones.
The following parameters are available for a cross-zone sharding configuration in thebinarystore.xmlfile:
readBehavior |
This parameter dictates the strategy for reading binaries from the mounts that make up the cross-zone sharded filestore. zone:BInaries are read from each mount according to zone settings. |
writeBehavior |
This parameter dictates the strategy for writing binaries to cross-zone sharding mounts: Possible values are: zonePercentageFreeSpace:Binaries are written to the mount in the relevant zone with the highest percentage of free space available. zoneFreeSpace:Binaries are written to the mount in the zone with the greatest absolute volume of free space available. |
Add to the ha-node.properties file
The following parameters are available for a cross-zone sharding configuration in theha-node.propertiesfile:
node.id |
Unique descriptive name of this server. Uniqueness Make sure that each node has an id that is unique on your whole network. |
cross.zone.order |
This parameter sets the zone order in which the data is written to the mounts. In the following example, "cross.zone.order=”us-east1,us-east2”,分片写入US-EAST-1区和then to the US-EAST-2 zone. |
You can dynamically add nodes to an existing sharding cluster using the ha-node.properties file. To do so, you will need your cluster to already be configured with sharding, and by adding the ‘cross.zone.order=us-east-1,us-east-2’property, the new node will be able to write to the existing cluster nodes without changing the binarystore.xml file.
Example:
This example displays a cross-zone sharding scenario in which the Artifactory cluster is configured with a redundancy of 2 and includes the following steps:
- The developer first deploys the package to the closest Artifactory node.
- The package is then automatically deployed to the 'US-EAST-1" zone to the shard with the highest percentage of free space in the "S1" shard (with 51% free space).
- The package is deployed using the same method to the "S3" shard, that also has the highest percentage of free space in the 'US-EAST-2' zone.
The code snippet below is a sample configuration of our cross-zone setup:
- 1 Artifactory cluster across 2 zones: "us-east-1" and "us-east-2" in this order.
- 4 HA nodes, 2 nodes in each zone.
- 4 mounts (shards), 2 mounts in each zone.
- The write strategy for the provider iszonePercentageFreeSpace.
Example: Cross-zone sharding configuration in ha-node.properties
node.id=west-node-1 cross.zone.order=”us-east-1,us-east-2”
Example: Cross-zone sharding configuration in the binarystore.xml
2 zone zonePercentageFreeSpace mount1 us-east-1 mount2 us-east-1 mount3 us-east-2 mount4 us-east-2 - 1 Artifactory cluster across 2 zones: "us-east-1" and "us-east-2" in this order.
Using Balancing to Recover from Mount Failure
In case of a mount failure, the actual redundancy in your system will be reduced accordingly. In the meantime, binaries continue to be written to the remaining active mounts. Once the malfunctioning mount has been restored, the system needs to rebalance the binaries written to the remaining active mounts to fully restore (i.e. balance) the redundancy configured in the system. Depending on how long the failed mount was inactive, this may involve a significant volume of binaries that now need to be written to the restored mount, which may take significant amount of time. Since restoring the full redundancy is a resource intensive operation, the balancing operation is run in a series of distinct sessions until complete. These are automatically invoked after aGarbage Collectionprocess has been run in the system.
Restoring Balance in Unbalanced Redundant Storage Units
在自愿的情况下行动,导致海事局alance the system redundancy, such as when doing a filestore migration, you may manually invoke rebalancing of redundancy using theOptimize System StorageREST API endpoint. Applying this endpoint raises a flag for Artifactory to run rebalancing following the next Garbage Collection. Note that, to expedite rebalancing, you can invoke garbage collection manually from the Artifactory UI.
Optimizing System Storage
Artifactory REST API provides an endpoint that allows you to raise a flag to indicate that Artifactory should invoke balancing between redundant storage units of a sharded filestore after the next garbage collection. For details, please refer toOptimize System Storage.
