Learn about the collection, configuration, and performance limits that apply when you import OSS data into Simple Log Service.
Collection limits
| Item | Description |
| Maximum object size |
Objects that exceed the size limit are skipped entirely. |
| Maximum record size | 3 MB per record. Records that exceed this limit are dropped. Dropped records appear as write failures on the Data Processing Insight dashboard. Related operations. |
| Object update behavior | Appending content to a previously imported object triggers a full re-import of that object. |
| New object detection interval | Minimum detection interval: 1 minute. Actual detection time may increase when many objects are pending import. |
Configuration limits
| Item | Description |
| Import configurations per project | Up to 100 import configurations of all types per project. To increase this quota, submit a ticket. |
Performance limits
| Item | Description |
| Concurrent subtasks | Each import configuration runs up to 8 concurrent subtasks by default. Each subtask processes up to 10 MB/s of decompressed data, for a total throughput of up to 80 MB/s per import task. To increase this quota, submit a ticket. |
| Logstore shard count | Write performance depends on the number of shards in the destination Logstore. Each shard supports a write throughput of 5 MB/s. For high-volume imports, increase the shard count. Manage shards. |
| Archive object restoration | Archive objects must be restored before import. Restoration typically takes about 1 minute. |
| Object size and throughput | Larger objects yield higher read throughput. Many small objects reduce throughput for the same total data volume. |
| Network and region | Place the OSS bucket and SLS project in the same region to minimize network costs and maximize transfer speed. Cross-region imports have lower read performance due to network latency. |
| New data import latency | Without OSS Metadata Indexing enabled, new object detection may exceed the configured check interval when many existing objects are present. For example, with approximately 1 million existing objects, detection latency is about 2 minutes. Latency scales linearly with the number of existing objects. |