From a3ada63fc70a60a7d5f115e24c2153a5f1362cdc Mon Sep 17 00:00:00 2001
From: Akashdeep Goel <akashdeepg@netflix.com>
Date: Mon, 27 May 2024 02:20:47 -0700
Subject: [PATCH] Adds support for disk_io in evcache

---
 service_capacity_modeling/models/common.py      | 17 +++++++++++++++++
 .../models/org/netflix/evcache.py               | 16 ++++++++++++++++
 2 files changed, 33 insertions(+)

diff --git a/service_capacity_modeling/models/common.py b/service_capacity_modeling/models/common.py
index 5ab5498..e8603a1 100644
--- a/service_capacity_modeling/models/common.py
+++ b/service_capacity_modeling/models/common.py
@@ -229,7 +229,15 @@ def compute_stateful_zone(
     # Some stateful clusters have preferences on per zone sizing
     cluster_size: Callable[[int], int] = lambda x: x,
     min_count: int = 0,
+    adjusted_disk_io_needed: Optional[float] = 0.0,
+    read_write_ratio: Optional[float] = 0.0
 ) -> ZoneClusterCapacity:
+
+
+    #If only reads, then max burst is X reads/s
+    #If only writes, then max burst is Y writes/s
+
+
     # Normalize the cores of this instance type to the latency reference
     needed_cores = math.ceil(
         max(1, needed_cores // (instance.cpu_ghz / core_reference_ghz))
@@ -269,6 +277,15 @@ def compute_stateful_zone(
     ):
         disk_per_node = min(max_local_disk_gib, instance.drive.size_gib)
         count = max(count, math.ceil(needed_disk_gib / disk_per_node))
+        if adjusted_disk_io_needed != 0.0:
+            instance_read_iops = drive.read_io_per_s
+            instance_write_iops = drive.write_io_per_s
+            instance_adjusted_io = (read_write_ratio * instance_read_iops + \
+                (1.0 - read_write_ratio) * instance_write_iops) * \
+                drive.block_size_kib * 1024
+            if instance_adjusted_io != 0.0:
+                count = max(count, math.ceil(adjusted_disk_io_needed / instance_adjusted_io))
+
 
     count = max(cluster_size(count), min_count)
     cost = count * instance.annual_cost
diff --git a/service_capacity_modeling/models/org/netflix/evcache.py b/service_capacity_modeling/models/org/netflix/evcache.py
index 41b3bbf..b0b6c92 100644
--- a/service_capacity_modeling/models/org/netflix/evcache.py
+++ b/service_capacity_modeling/models/org/netflix/evcache.py
@@ -232,6 +232,20 @@ def reserve_memory(instance_mem_gib):
     if desires.service_tier < 1:
         min_count = 2
 
+    is_disk_io_constraint: bool = requirement.disk_gib > 0.0
+    adjusted_disk_io_needed = 0.0
+    read_write_ratio = 0.0
+    if is_disk_io_constraint:
+        reads_per_sec = desires.query_pattern.estimated_read_per_second.mid
+        writes_per_sec = desires.query_pattern.estimated_write_per_second.mid
+        read_size = desires.query_pattern.estimated_mean_read_size_bytes.mid
+        write_size = desires.query_pattern.estimated_mean_write_size_bytes.mid
+        read_disk_io_needed = reads_per_sec * read_size
+        write_disk_io_needed = writes_per_sec * write_size
+        adjusted_disk_io_needed = (reads_per_sec * read_disk_io_needed + writes_per_sec * write_disk_io_needed) / \
+            (reads_per_sec + writes_per_sec)
+        read_write_ratio = reads_per_sec / (reads_per_sec + writes_per_sec)
+
     cluster = compute_stateful_zone(
         instance=instance,
         drive=drive,
@@ -248,6 +262,8 @@ def reserve_memory(instance_mem_gib):
         # Sidecars and Variable OS Memory
         reserve_memory=lambda x: base_mem,
         core_reference_ghz=requirement.core_reference_ghz,
+        adjusted_disk_io_needed = adjusted_disk_io_needed,
+        read_write_ratio = read_write_ratio
     )
     # Communicate to the actual provision that if we want reduced RF
     params = {"evcache.copies": copies_per_region}