diff --git a/src/dexorder/bin/main.py b/src/dexorder/bin/main.py
index 1933516..d757109 100644
--- a/src/dexorder/bin/main.py
+++ b/src/dexorder/bin/main.py
@@ -117,7 +117,7 @@ async def main():
             if redis_state:
                 # load initial state
                 log.info('initializing redis with root state')
-                await redis_state.save(state.root_fork, state.diffs_by_branch[state.root_branch.id])
+                await redis_state.init(state, state.root_fork)
 
     await initialize_accounting_runner()
 
diff --git a/src/dexorder/memcache/__init__.py b/src/dexorder/memcache/__init__.py
index 9f13988..56bc6c3 100644
--- a/src/dexorder/memcache/__init__.py
+++ b/src/dexorder/memcache/__init__.py
@@ -4,45 +4,18 @@ from contextvars import ContextVar
 
 import redis.asyncio as redis_async
 from redis.asyncio import Redis
-from redis.asyncio.client import Pipeline
 
 from dexorder import config
 
 log = logging.getLogger(__name__)
 
 
-class FlushingPipeline:
-    def __init__(self, redis: Redis):
-        self.redis = redis
-        self.pipe: Pipeline = redis.pipeline()
-        self.full_pipes: list[Pipeline] = []
-        self.count = 0
-        self.flush_at = 10_000
-
-    def __getattr__(self, item):
-        if item in ('sadd', 'srem', 'hset', 'hdel', 'json'):
-            self.count += 1
-            if self.count >= self.flush_at:
-                self.full_pipes.append(self.pipe)
-                self.pipe = self.redis.pipeline()
-                self.count = 0
-        return getattr(self.pipe, item)
-
-    async def execute(self):
-        for pipe in self.full_pipes:
-            await pipe.execute()
-        await self.pipe.execute()
-        self.pipe = None
-        self.full_pipes.clear()
-        self.count = 0
-
-
 class Memcache:
     @staticmethod
     @asynccontextmanager
-    async def batch():
+    async def batch(transaction=True):
         old_redis: Redis = current_redis.get()
-        pipe = FlushingPipeline(old_redis)
+        pipe = old_redis.pipeline(transaction=transaction)
         # noinspection PyTypeChecker
         current_redis.set(pipe)
         try:
diff --git a/src/dexorder/memcache/memcache_state.py b/src/dexorder/memcache/memcache_state.py
index 9aa0bdb..b3b2faf 100644
--- a/src/dexorder/memcache/memcache_state.py
+++ b/src/dexorder/memcache/memcache_state.py
@@ -2,7 +2,7 @@ import logging
 from collections import defaultdict
 from typing import Iterable, Union, Reversible, Any
 
-from redis.asyncio.client import Pipeline, Redis
+from redis.asyncio.client import Pipeline
 from socket_io_emitter import Emitter
 
 from dexorder import DELETE
@@ -40,11 +40,12 @@ class RedisState (SeriesCollection):
         for series in self.datas.keys():
             for k, v in state.iteritems(fork, series):
                 diffs.append(DiffItem(series, k, v))
-        await self.save(fork, diffs)
+        # todo tim fix pubs
+        await self.save(fork, diffs, use_transaction=True, skip_pubs=True)  # use_transaction=False if the data is too big
 
 
     # noinspection PyAsyncCall
-    async def save(self, fork: Fork, diffs: Reversible[Union[DiffItem, DiffEntryItem]]):
+    async def save(self, fork: Fork, diffs: Reversible[Union[DiffItem, DiffEntryItem]], *, use_transaction=True, skip_pubs=False):
         # the diffs must be already compressed such that there is only one action per key
         chain = current_chain.get()
         chain_id = chain.id
@@ -91,7 +92,9 @@ class RedisState (SeriesCollection):
                     hsets[series][key] = value
                 else:
                     raise NotImplementedError
-        async with memcache.batch() as r:
+        async with memcache.batch(use_transaction) as r:
+            # Redis pipelines fill up before our state can be sent, so we cannot do this atomically.
+            # However, sending many individual calls is super slow, so we
             r: Pipeline
             for series, keys in sadds.items():
                 r.sadd(series, *keys)
@@ -106,7 +109,7 @@ class RedisState (SeriesCollection):
             r.json(json_encoder).set(block_series,'$',[fork.height, headstr])
             pubs.append((str(chain_id), 'head', [fork.height, headstr]))
         # separate batch for pubs
-        if pubs:
+        if pubs and not skip_pubs:
             await publish_all(pubs)