diff --git a/continuous_integration/simple_test.sh b/continuous_integration/simple_test.sh
index 5fe39ac..74980f3 100644
--- a/continuous_integration/simple_test.sh
+++ b/continuous_integration/simple_test.sh
@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 source activate test
-conda install -c conda-forge -y lxml py4j
+conda install -c conda-forge -y py4j hdfs3
 cd /knit
 python setup.py install mvn
 py.test -vv
\ No newline at end of file
diff --git a/knit/core.py b/knit/core.py
index 949e896..b3b4d23 100644
--- a/knit/core.py
+++ b/knit/core.py
@@ -18,6 +18,7 @@
 from .env import CondaCreator
 from .exceptions import KnitException, YARNException
 from .yarn_api import YARNAPI
+from .utils import triple_slash
 
 from py4j.protocol import Py4JError
 from py4j.java_gateway import JavaGateway, GatewayClient
@@ -313,12 +314,15 @@ def preexec_func():
                             " Check that java is installed and the Knit JAR"
                             " file exists.")
 
-        gateway = JavaGateway(GatewayClient(port=gateway_port), auto_convert=True)
+        gateway = JavaGateway(GatewayClient(port=gateway_port),
+                              auto_convert=True)
         self.client = gateway.entry_point
         self.client_gateway = gateway
-        files = [(f if self.check_needs_upload(f) else ('hdfs://' + f))
-                 for f in files]
-        jfiles = ListConverter().convert(files, gateway._gateway_client)
+        logger.debug("Files submitted: %s" % files)
+        upfiles = [f for f in files if (not f.startswith('hdfs://')
+                   and self.check_needs_upload(f))]
+        logger.debug("Files to upload: %s" % upfiles)
+        jfiles = ListConverter().convert(upfiles, gateway._gateway_client)
         jenv = MapConverter().convert(envvars, gateway._gateway_client)
 
         self.app_id = self.client.start(jfiles, jenv, app_name, queue)
@@ -347,7 +351,11 @@ def preexec_func():
         gateway = JavaGateway(GatewayClient(
             address=master_rpchost, port=master_rpcport), auto_convert=True)
         self.master = gateway.entry_point
-        jfiles = ListConverter().convert(files, gateway._gateway_client)
+        rfiles = [triple_slash(f) if f.startswith('hdfs://') else
+                  '/'.join([self.hdfs_home, '.knitDeps', os.path.basename(f)])
+                  for f in files]
+        logger.debug("Resource files: %s" % rfiles)
+        jfiles = ListConverter().convert(rfiles, gateway._gateway_client)
         jenv = MapConverter().convert(envvars, gateway._gateway_client)
         self.master.init(jfiles, jenv, cmd, num_containers,
                          virtual_cores, memory)
@@ -585,7 +593,7 @@ def check_needs_upload(self, path):
         """Upload is needed if file does not exist in HDFS or is older"""
         if self.upload_always:
             return True
-        fn = (self.hdfs_home + '/.knitDeps/' + os.path.basename(path))
+        fn = '/'.join([self.hdfs_home, '.knitDeps', os.path.basename(path)])
         if self.hdfs and self.hdfs.exists(fn):
             st = os.stat(path)
             size = st.st_size
diff --git a/knit/dask_yarn.py b/knit/dask_yarn.py
index 0452572..9c1217a 100644
--- a/knit/dask_yarn.py
+++ b/knit/dask_yarn.py
@@ -29,7 +29,10 @@ class DaskYARNCluster(object):
     ----------
     nn, nn_port, rm, rm_port, user, autodetect, lang: see knit.Knit
     env: str or None
-        If provided, the path of a zipped conda env to put in containers
+        If provided, the path of a zipped conda env to put in containers. This
+        can be a local zip file to upload, a zip file already on HDFS (hdfs://)
+        or a directory to zip and upload. If not provided, a default environment
+        will be built, containing dask.
     packages: list of str
         Packages to install in the env to provide to containers *if* env is 
         None. Uses conda spec for pinning versions. dask and distributed will
@@ -111,7 +114,8 @@ def start(self, n_workers=1, cpus=1, memory=2048, checks=True,
                 # create env from scratch
                 self.env = c.create_env(env_name=env_name,
                                         packages=self.packages)
-        elif not self.env.endswith('.zip'):
+        elif (not self.env.endswith('.zip') and
+              not self.env.startswith('hdfs://')):
             # given env directory, so zip it
             self.env = zip_path(self.env)
 
@@ -124,7 +128,8 @@ def start(self, n_workers=1, cpus=1, memory=2048, checks=True,
                    ''.format(cpus=cpus, mem=memory * 1e6, pref=pref,
                              addr=self.local_cluster.scheduler.address))
 
-        app_id = self.knit.start(command, files=[self.env],
+        files = [self.env] + kwargs.pop('files', [])
+        app_id = self.knit.start(command, files=files,
                                  num_containers=n_workers, virtual_cores=cpus,
                                  memory=memory, checks=checks, **kwargs)
         self.app_id = app_id
diff --git a/knit/tests/test_core.py b/knit/tests/test_core.py
index e17ffaf..ac0d9ef 100644
--- a/knit/tests/test_core.py
+++ b/knit/tests/test_core.py
@@ -208,10 +208,10 @@ def test_hdfs_home():
     try:
         hdfs.mkdir(d)
         k = Knit(nn='localhost', rm='localhost', nn_port=8020, rm_port=8088,
-               replication_factor=1, hdfs_home=d)
+                 replication_factor=1, hdfs_home=d)
 
         env_zip = k.create_env(env_name='dev', packages=['python=2.7'], remove=True)
-        k.start('env', env=env_zip, memory=128)
+        k.start('env', files=[env_zip], memory=128)
 
         assert d + '/.knitDeps' in hdfs.ls(d, False)
         assert d + "/.knitDeps/knit-1.0-SNAPSHOT.jar" in hdfs.ls(d + '/.knitDeps', False)
@@ -219,10 +219,6 @@ def test_hdfs_home():
         if not k.wait_for_completion(30):
             k.kill()
 
-        time.sleep(5)  # log aggregation
-        logs = k.logs()
-        assert "PYTHON_BIN=" in str(logs)
-        assert "CONDA_PREFIX=" in str(logs)
     finally:
         hdfs.rm(d, True)
         k.kill()
@@ -338,3 +334,17 @@ def test_kill_all(k):
     k.yarn_api.kill_all()
     time.sleep(2)
     assert k.runtime_status() == 'KILLED'
+
+
+def test_existing_path(k):
+    # TODO: is this a good test if we noisily log the upload of files?
+    hdfs3 = pytest.importorskip('hdfs3')
+    hdfs = hdfs3.HDFileSystem()
+    k.hdfs = hdfs
+    cmd = 'ls -l'
+    hdfs.put(__file__, '/tmp/mytestfile')
+    k.start(cmd, files=['hdfs://tmp/mytestfile'])
+    wait_for_status(k, 'FINISHED')
+    time.sleep(2)
+    out = k.logs()
+    assert ' mytestfile ' in str(out)
diff --git a/knit/tests/test_dask.py b/knit/tests/test_dask.py
index a1873f6..9a25124 100644
--- a/knit/tests/test_dask.py
+++ b/knit/tests/test_dask.py
@@ -110,7 +110,7 @@ def test_yarn_cluster_add_stop(loop):
 
     client = Client(cluster)
     future = client.submit(lambda x: x + 1, 10)
-    assert future.result() == 11
+    assert future.result() == 11    # waits for cluster to have a worker
 
     info = client.scheduler_info()
     workers = info['workers']
@@ -120,37 +120,28 @@ def test_yarn_cluster_add_stop(loop):
     num_containers = status['runningContainers']
     assert num_containers == 2  # 1 container for the worker and 1 for the AM
 
+    # Add a worker
     cluster.add_workers(n_workers=1, cpus=1, memory=128)
-
-    while num_containers != 3:
-        status = cluster.knit.status()
-        num_containers = status['runningContainers']
-
-    # wait a tad to let workers connect to scheduler
-
     start = time.time()
     while len(client.scheduler_info()['workers']) < 2:
-        time.sleep(0.1)
-        assert time.time() < start + 10
+        time.sleep(1)
+        assert time.time() < start + 60
 
+    status = cluster.knit.status()
+    num_containers = status['runningContainers']
     assert num_containers == 3
-    info = client.scheduler_info()
-    workers = info['workers']
-    assert len(workers) == 2
-
     assert len(cluster.workers) == 2
 
+    # Remove a worker
     cluster.remove_worker(cluster.workers[1])
-    while num_containers != 2:
-        status = cluster.knit.status()
-        num_containers = status['runningContainers']
-
-    time.sleep(2)
-    assert len(cluster.workers) == 1
+    start = time.time()
+    while len(client.scheduler_info()['workers']) == 2:
+        time.sleep(1)
+        assert time.time() < start + 10
 
     # STOP ALL WORKERS!
     cluster.stop()
-    time.sleep(2)
-
-    workers = client.scheduler_info()['workers']
-    assert len(workers) == 0
+    start = time.time()
+    while len(client.scheduler_info()['workers']) > 0:
+        time.sleep(2)
+        assert time.time() < start + 10
diff --git a/knit/tests/test_utils.py b/knit/tests/test_utils.py
index 4b49841..38bd352 100644
--- a/knit/tests/test_utils.py
+++ b/knit/tests/test_utils.py
@@ -1,6 +1,6 @@
 import logging
 
-from knit.utils import set_logging
+from knit.utils import set_logging, triple_slash
 
 
 def test_set_logging():
@@ -9,3 +9,9 @@ def test_set_logging():
     assert logger.level == logging.DEBUG
     set_logging(logging.INFO)
     assert logger.level == logging.INFO
+
+
+def test_slashes():
+    assert triple_slash('hdfs://hello/path') == 'hdfs:///hello/path'
+    assert triple_slash('hdfs:///hello/path') == 'hdfs:///hello/path'
+    assert triple_slash('hdfs:////hello/path') == 'hdfs:////hello/path'
diff --git a/knit/utils.py b/knit/utils.py
index 8b951ee..ff0bec6 100644
--- a/knit/utils.py
+++ b/knit/utils.py
@@ -43,3 +43,10 @@ def get_log_content(s):
     ind1 = s[ind0:].find("</td>")
     out = s[ind0:ind0+ind1]
     return out.lstrip('\n          <pre>').rstrip('</pre>\n        ')
+
+
+def triple_slash(s):
+    if s.startswith('hdfs://') and not s.startswith('hdfs:///'):
+        return 'hdfs:///' + s[7:]
+    else:
+        return s
diff --git a/knit_jvm/src/main/scala/io/continuum/knit/ApplicationMaster.scala b/knit_jvm/src/main/scala/io/continuum/knit/ApplicationMaster.scala
index 876fc7f..d67e279 100644
--- a/knit_jvm/src/main/scala/io/continuum/knit/ApplicationMaster.scala
+++ b/knit_jvm/src/main/scala/io/continuum/knit/ApplicationMaster.scala
@@ -193,9 +193,8 @@ object ApplicationMaster extends Logging with AMRMClientAsync.CallbackHandler wi
             val fileUpload = Records.newRecord(classOf[LocalResource])
             var p = new Path(fileName)
             val name = p.getName  
-            val p2 = new Path(".knitDeps/" + name)
-            logger.info(f"RESOURCE: $p => $p2 archive=$iszip")
-            setUpLocalResource(p2, fileUpload, archived=iszip)
+            logger.info(f"RESOURCE: $p archive=$iszip")
+            setUpLocalResource(p, fileUpload, archived=iszip)
             localResources(name) = fileUpload
           }
         }