service-capacity-modeling 0.3.53__py3-none-any.whl → 0.3.55__py3-none-any.whl

service_capacity_modeling/capacity_planner.py

@@ -511,6 +511,7 @@ class CapacityPlanner:
         num_results: Optional[int] = None,
         num_regions: int = 3,
         extra_model_arguments: Optional[Dict[str, Any]] = None,
+        max_results_per_family: int = 1,
     ) -> Sequence[CapacityPlan]:
         if model_name not in self._models:
             raise ValueError(
@@ -538,6 +539,7 @@ class CapacityPlanner:
                 lifecycles=lifecycles,
                 instance_families=instance_families,
                 drives=drives,
+                max_results_per_family=max_results_per_family,
             )
             if sub_plan:
                 results.append(sub_plan)
@@ -555,6 +557,7 @@ class CapacityPlanner:
         instance_families: Optional[Sequence[str]] = None,
         drives: Optional[Sequence[str]] = None,
         extra_model_arguments: Optional[Dict[str, Any]] = None,
+        max_results_per_family: int = 1,
     ) -> Sequence[CapacityPlan]:
         extra_model_arguments = extra_model_arguments or {}
         model = self._models[model_name]
@@ -577,7 +580,9 @@ class CapacityPlanner:
         plans.sort(key=lambda p: (p.rank, p.candidate_clusters.total_annual_cost))
 
         num_results = num_results or self._default_num_results
-        return reduce_by_family(plans)[:num_results]
+        return reduce_by_family(plans, max_results_per_family=max_results_per_family)[
+            :num_results
+        ]
 
     # Calculates the minimum cpu, memory, and network requirements based on desires.
     def _per_instance_requirements(self, desires) -> Tuple[int, float]:
@@ -696,6 +701,7 @@ class CapacityPlanner:
         regret_params: Optional[CapacityRegretParameters] = None,
         extra_model_arguments: Optional[Dict[str, Any]] = None,
         explain: bool = False,
+        max_results_per_family: int = 1,
     ) -> UncertainCapacityPlan:
         extra_model_arguments = extra_model_arguments or {}
 
@@ -741,6 +747,7 @@ class CapacityPlanner:
                             lifecycles=lifecycles,
                             instance_families=instance_families,
                             drives=drives,
+                            max_results_per_family=max_results_per_family,
                         ),
                     )
                 )
@@ -763,7 +770,8 @@ class CapacityPlanner:
                 ],
                 zonal_requirements,
                 regional_requirements,
-            )
+            ),
+            max_results_per_family=max_results_per_family,
         )[:num_results]
 
         low_p, high_p = sorted(percentiles)[0], sorted(percentiles)[-1]

service_capacity_modeling/models/org/netflix/cassandra.py

@@ -144,6 +144,7 @@ def _estimate_cassandra_requirement(  # pylint: disable=too-many-positional-argu
     disk_buffer = buffer_for_components(
         buffers=desires.buffers, components=[BufferComponent.disk]
     )
+    memory_preserve = False
     reference_shape = desires.reference_shape
     current_capacity = (
         None
@@ -169,6 +170,9 @@ def _estimate_cassandra_requirement(  # pylint: disable=too-many-positional-argu
             * current_capacity.cluster_instance_count.mid
             * (disk_scale or 1)
         )
+        _, memory_preserve = derived_buffer_for_component(
+            desires.buffers.derived, ["storage", "memory"]
+        )
     else:
         # If the cluster is not yet provisioned
         capacity_requirement = _zonal_requirement_for_new_cluster(
@@ -202,13 +206,19 @@ def _estimate_cassandra_requirement(  # pylint: disable=too-many-positional-argu
     )
     rps_working_set = min(1.0, disk_rps / max_rps_to_disk)
 
-    # If disk RPS will be smaller than our target because there are no
-    # reads, we don't need to hold as much data in memory.
-    # For c*, we can skip memory buffer and can just keep using the heap and write buffer calc
-    # Eventually we'll want to phrase those heap, read cache, and write cache as buffers
-    needed_memory = min(working_set, rps_working_set) * disk_used_gib * zones_per_region
-    # Now convert to per zone
-    needed_memory = max(1, int(needed_memory // zones_per_region))
+    if memory_preserve:
+        needed_memory = capacity_requirement.mem_gib.mid
+    else:
+        # If disk RPS will be smaller than our target because there are no
+        # reads, we don't need to hold as much data in memory.
+        # For c*, we can skip memory buffer and can just keep using the heap and write buffer calc
+        # Eventually we'll want to phrase those heap, read cache, and write cache as buffers
+        needed_memory = (
+            min(working_set, rps_working_set) * disk_used_gib * zones_per_region
+        )
+        # Now convert to per zone
+        needed_memory = max(1, int(needed_memory // zones_per_region))
+
     logger.debug(
         "Need (cpu, mem, disk, working) = (%s, %s, %s, %f)",
         needed_cores,
@@ -381,6 +391,7 @@ def _estimate_cassandra_cluster_zonal(  # pylint: disable=too-many-positional-ar
         # C* clusters provision in powers of 2 because doubling
         cluster_size=next_power_of_2,
         min_count=max(min_count, required_cluster_size or 0),
+        # TODO: Take reserve memory calculation into account during buffer calculation
         # C* heap usage takes away from OS page cache memory
         reserve_memory=lambda x: base_mem + heap_fn(x),
         # C* heap buffers the writes at roughly a rate of

service_capacity_modeling/models/utils.py

@@ -1,19 +1,26 @@
 import math
+from typing import Dict
 from typing import Iterable
 from typing import List
-from typing import Set
 from typing import Tuple
 
 from service_capacity_modeling.models import CapacityPlan
 
 
-def reduce_by_family(plans: Iterable[CapacityPlan]) -> List[CapacityPlan]:
+def reduce_by_family(
+    plans: Iterable[CapacityPlan], max_results_per_family: int = 1
+) -> List[CapacityPlan]:
     """Groups a potential set of clusters by hardware family sorted by cost.
 
     Useful for showing different family options.
+
+    Args:
+        plans: Iterable of CapacityPlan objects to filter
+        max_results_per_family: Maximum number of results to return per
+            family combination
     """
-    zonal_families: Set[Tuple[Tuple[str, str], ...]] = set()
-    regional_families: Set[Tuple[Tuple[str, str], ...]] = set()
+    zonal_families: Dict[Tuple[Tuple[str, str], ...], int] = {}
+    regional_families: Dict[Tuple[Tuple[str, str], ...], int] = {}
 
     result: List[CapacityPlan] = []
     for plan in plans:
@@ -31,11 +38,20 @@ def reduce_by_family(plans: Iterable[CapacityPlan]) -> List[CapacityPlan]:
                 sorted({(c.cluster_type, c.instance.family) for c in topo.zonal})
             )
 
-        if not (zonal_type in zonal_families and regional_type in regional_families):
+        # Count how many of each family combination we've seen
+        zonal_count = zonal_families.get(zonal_type, 0)
+        regional_count = regional_families.get(regional_type, 0)
+
+        # Add the plan if we haven't reached the maximum for either family type
+        if (
+            zonal_count < max_results_per_family
+            or regional_count < max_results_per_family
+        ):
             result.append(plan)
 
-        regional_families.add(regional_type)
-        zonal_families.add(zonal_type)
+            # Update counters
+            zonal_families[zonal_type] = zonal_count + 1
+            regional_families[regional_type] = regional_count + 1
 
     return result
 

{service_capacity_modeling-0.3.53.dist-info → service_capacity_modeling-0.3.55.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: service-capacity-modeling
-Version: 0.3.53
+Version: 0.3.55
 Summary: Contains utilities for modeling capacity for pluggable workloads
 Author: Joseph Lynch
 Author-email: josephl@netflix.com

{service_capacity_modeling-0.3.53.dist-info → service_capacity_modeling-0.3.55.dist-info}/RECORD

@@ -1,5 +1,5 @@
 service_capacity_modeling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-service_capacity_modeling/capacity_planner.py,sha256=0lQCQRlnEF2frwuK0cOy5wAkOcr1-6qunBjZPukTcdg,31525
+service_capacity_modeling/capacity_planner.py,sha256=B-C6EgZEWLKS6HEL74WxGeoLHrzNN6Cmyfee2jKsdcU,31915
 service_capacity_modeling/interface.py,sha256=145YV4Lm2YZhBAKjsdUIjeIOxJ7NxqUPwz9oxekgsPo,36995
 service_capacity_modeling/stats.py,sha256=8HIPwVnmvbauBwXhn6vbNYO7-CzWPuymnq0eX7ZA1_w,5849
 service_capacity_modeling/hardware/__init__.py,sha256=Lc-DdX2ZJ_sFT3KlImPUUcKmcjVnavH_mMXlyCwpNiw,9001
@@ -42,11 +42,11 @@ service_capacity_modeling/hardware/profiles/shapes/aws/manual_services.json,sha2
 service_capacity_modeling/models/__init__.py,sha256=XK7rTBW8ZXQY5L9Uy2FwjuFN_KBW3hKw7IrhG1piajs,13567
 service_capacity_modeling/models/common.py,sha256=ixmNkohvw30R2JBvdpM6xKb19na_g3GNGvQQYLlkV_A,33577
 service_capacity_modeling/models/headroom_strategy.py,sha256=QIkP_K_tK2EGAjloaGfXeAPH5M0UDCN8FlAtwV9xxTA,651
-service_capacity_modeling/models/utils.py,sha256=Y9xwSzLuGQ7Tn4BCkbdMtR7_fjEQ9PKQ52JlHVtdEtY,1567
+service_capacity_modeling/models/utils.py,sha256=0F__wz9KAGhPIQfvNp-FTtTANW6-sO4FsyddnuXqSJc,2161
 service_capacity_modeling/models/org/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 service_capacity_modeling/models/org/netflix/__init__.py,sha256=m7IaQbo85NEbDvfoPJREIznpzg0YHTCrKP5C1GvnOYM,2378
 service_capacity_modeling/models/org/netflix/aurora.py,sha256=Mi9zd48k64GkKIjAs3J1S2qThguNvyWIy2dUmhwrVhc,12883
-service_capacity_modeling/models/org/netflix/cassandra.py,sha256=kJ68ufYTo849Zsh6_BXn-mSLse0ZHsMEP0yuzeFrTzE,32509
+service_capacity_modeling/models/org/netflix/cassandra.py,sha256=mglxXmEFF_ge8_HD02mHuFp4leLEA4JF3xDAS1h7bW4,32896
 service_capacity_modeling/models/org/netflix/counter.py,sha256=hOVRRCgCPU-A5TdLKQXc_mWTQpkKOWRNjOeECdDP7kA,9205
 service_capacity_modeling/models/org/netflix/crdb.py,sha256=2rD4Io0yT7o0NR4lNferXXOSTDe0SkT1LbSChvNgRrQ,19698
 service_capacity_modeling/models/org/netflix/ddb.py,sha256=2jxMFz31xckJvymvVlu1yWm0X4dGYlqxDo0bftU1B9M,26307
@@ -69,9 +69,9 @@ service_capacity_modeling/tools/auto_shape.py,sha256=Rk5Fjrw2susVL8It_J2KUADoMGB
 service_capacity_modeling/tools/fetch_pricing.py,sha256=SHOtFaPr61op2bnY9i_g_1-d-Nz2rV8c7Jwsye2R49s,3763
 service_capacity_modeling/tools/generate_missing.py,sha256=WN7Y0iPjelJu9d2Yq5dcu_EBIQPX2VxKgOou4NMNOM4,2860
 service_capacity_modeling/tools/instance_families.py,sha256=e9JWSIdljSmHI8Nb2MI5Ld9JqQ7WdOtPtV7g3oR7ZiU,7764
-service_capacity_modeling-0.3.53.dist-info/licenses/LICENSE,sha256=nl_Lt5v9VvJ-5lWJDT4ddKAG-VZ-2IaLmbzpgYDz2hU,11343
-service_capacity_modeling-0.3.53.dist-info/METADATA,sha256=Bip_YLLrgI7oTgIctl9rxmgxdqVr0DuQycV8lzOsDWU,9590
-service_capacity_modeling-0.3.53.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-service_capacity_modeling-0.3.53.dist-info/entry_points.txt,sha256=ZsjzpG5SomWpT1zCE19n1uSXKH2gTI_yc33sdl0vmJg,146
-service_capacity_modeling-0.3.53.dist-info/top_level.txt,sha256=H8XjTCLgR3enHq5t3bIbxt9SeUkUT8HT_SDv2dgIT_A,26
-service_capacity_modeling-0.3.53.dist-info/RECORD,,
+service_capacity_modeling-0.3.55.dist-info/licenses/LICENSE,sha256=nl_Lt5v9VvJ-5lWJDT4ddKAG-VZ-2IaLmbzpgYDz2hU,11343
+service_capacity_modeling-0.3.55.dist-info/METADATA,sha256=E-VMoK90ERxd_OFWoijTlHR-KPdrcbM9v1SMzdSaUDM,9590
+service_capacity_modeling-0.3.55.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+service_capacity_modeling-0.3.55.dist-info/entry_points.txt,sha256=ZsjzpG5SomWpT1zCE19n1uSXKH2gTI_yc33sdl0vmJg,146
+service_capacity_modeling-0.3.55.dist-info/top_level.txt,sha256=H8XjTCLgR3enHq5t3bIbxt9SeUkUT8HT_SDv2dgIT_A,26
+service_capacity_modeling-0.3.55.dist-info/RECORD,,