`DEBUG POPULATE` OOM

[andydunstall]

Describe the bug Using DEBUG POPULATE on a master with one replica causes Dragonfly to run out of memory with high RSS

To Reproduce Created a master and replica with 200GB each using Dragonfly 1.16.1.

Running redis-cli -p 9999 debug populate 50000000 test 1000 rand to get 50GB of data worked ok.

Then running redis-cli -p 9999 debug populate 100000000 test 1000 rand to get 100GB of data (on a 200GB machine) and Dragonfly OOM'ed

Running an INFO memory while populate was running (prior to OOM) shows high RSS and replication_streaming_buffer_bytes:

wql1kr9g7.dev.dragonflydb.cloud:6385> info memory
# Memory
used_memory:81590059392
used_memory_human:75.99GiB
used_memory_peak:72276289920
used_memory_peak_human:67.31GiB
fibers_stack_vms:4159504
fibers_count:127
used_memory_rss:134092623872
used_memory_rss_human:124.88GiB
used_memory_peak_rss:134092623872
maxmemory:199999094784
maxmemory_human:186.26GiB
object_used_memory:77293060096
type_used_memory_STRING:77293060096
table_used_memory:4233477888
num_buckets:7864320
num_entries:86264576
inline_keys:86264576
listpack_blobs:0
listpack_bytes:0
small_string_bytes:77293060096
pipeline_cache_bytes:0
dispatch_queue_bytes:2016
dispatch_queue_subscriber_bytes:0
dispatch_queue_peak_bytes:6552
client_read_buffer_peak_bytes:18688
tls_bytes:0
cache_mode:store
maxmemory_policy:noeviction
replication_streaming_buffer_bytes:71403831296
replication_full_sync_buffer_bytes:0

Repeating without replication works ok.

Environment (please complete the following information):

• Instance: x2gd.4xlarge
• Dragonfly Version: v1.16.1