Show More
@@ -1,520 +1,545 b'' | |||||
1 | """ |
|
1 | """ | |
2 | Gunicorn config extension and hooks. This config file adds some extra settings and memory management. |
|
2 | Gunicorn config extension and hooks. This config file adds some extra settings and memory management. | |
3 | Gunicorn configuration should be managed by .ini files entries of RhodeCode or VCSServer |
|
3 | Gunicorn configuration should be managed by .ini files entries of RhodeCode or VCSServer | |
4 | """ |
|
4 | """ | |
5 |
|
5 | |||
6 | import gc |
|
6 | import gc | |
7 | import os |
|
7 | import os | |
8 | import sys |
|
8 | import sys | |
9 | import math |
|
9 | import math | |
10 | import time |
|
10 | import time | |
11 | import threading |
|
11 | import threading | |
12 | import traceback |
|
12 | import traceback | |
13 | import random |
|
13 | import random | |
14 | import socket |
|
14 | import socket | |
15 | import dataclasses |
|
15 | import dataclasses | |
|
16 | import json | |||
16 | from gunicorn.glogging import Logger |
|
17 | from gunicorn.glogging import Logger | |
17 |
|
18 | |||
18 |
|
19 | |||
19 | def get_workers(): |
|
20 | def get_workers(): | |
20 | import multiprocessing |
|
21 | import multiprocessing | |
21 | return multiprocessing.cpu_count() * 2 + 1 |
|
22 | return multiprocessing.cpu_count() * 2 + 1 | |
22 |
|
23 | |||
23 |
|
24 | |||
24 | bind = "127.0.0.1:10010" |
|
25 | bind = "127.0.0.1:10010" | |
25 |
|
26 | |||
26 |
|
27 | |||
27 | # Error logging output for gunicorn (-) is stdout |
|
28 | # Error logging output for gunicorn (-) is stdout | |
28 | errorlog = '-' |
|
29 | errorlog = '-' | |
29 |
|
30 | |||
30 | # Access logging output for gunicorn (-) is stdout |
|
31 | # Access logging output for gunicorn (-) is stdout | |
31 | accesslog = '-' |
|
32 | accesslog = '-' | |
32 |
|
33 | |||
33 |
|
34 | |||
34 | # SERVER MECHANICS |
|
35 | # SERVER MECHANICS | |
35 | # None == system temp dir |
|
36 | # None == system temp dir | |
36 | # worker_tmp_dir is recommended to be set to some tmpfs |
|
37 | # worker_tmp_dir is recommended to be set to some tmpfs | |
37 | worker_tmp_dir = None |
|
38 | worker_tmp_dir = None | |
38 | tmp_upload_dir = None |
|
39 | tmp_upload_dir = None | |
39 |
|
40 | |||
40 | # use re-use port logic |
|
41 | # use re-use port logic to let linux internals load-balance the requests better. | |
41 |
|
|
42 | reuse_port = True | |
42 |
|
43 | |||
43 | # Custom log format |
|
44 | # Custom log format | |
44 | #access_log_format = ( |
|
45 | #access_log_format = ( | |
45 | # '%(t)s %(p)s INFO [GNCRN] %(h)-15s rqt:%(L)s %(s)s %(b)-6s "%(m)s:%(U)s %(q)s" usr:%(u)s "%(f)s" "%(a)s"') |
|
46 | # '%(t)s %(p)s INFO [GNCRN] %(h)-15s rqt:%(L)s %(s)s %(b)-6s "%(m)s:%(U)s %(q)s" usr:%(u)s "%(f)s" "%(a)s"') | |
46 |
|
47 | |||
47 | # loki format for easier parsing in grafana |
|
48 | # loki format for easier parsing in grafana | |
48 | access_log_format = ( |
|
49 | loki_access_log_format = ( | |
49 | 'time="%(t)s" pid=%(p)s level="INFO" type="[GNCRN]" ip="%(h)-15s" rqt="%(L)s" response_code="%(s)s" response_bytes="%(b)-6s" uri="%(m)s:%(U)s %(q)s" user=":%(u)s" user_agent="%(a)s"') |
|
50 | 'time="%(t)s" pid=%(p)s level="INFO" type="[GNCRN]" ip="%(h)-15s" rqt="%(L)s" response_code="%(s)s" response_bytes="%(b)-6s" uri="%(m)s:%(U)s %(q)s" user=":%(u)s" user_agent="%(a)s"') | |
50 |
|
51 | |||
|
52 | # JSON format | |||
|
53 | json_access_log_format = json.dumps({ | |||
|
54 | 'time': r'%(t)s', | |||
|
55 | 'pid': r'%(p)s', | |||
|
56 | 'level': 'INFO', | |||
|
57 | 'ip': r'%(h)s', | |||
|
58 | 'request_time': r'%(L)s', | |||
|
59 | 'remote_address': r'%(h)s', | |||
|
60 | 'user_name': r'%(u)s', | |||
|
61 | 'status': r'%(s)s', | |||
|
62 | 'method': r'%(m)s', | |||
|
63 | 'url_path': r'%(U)s', | |||
|
64 | 'query_string': r'%(q)s', | |||
|
65 | 'protocol': r'%(H)s', | |||
|
66 | 'response_length': r'%(B)s', | |||
|
67 | 'referer': r'%(f)s', | |||
|
68 | 'user_agent': r'%(a)s', | |||
|
69 | ||||
|
70 | }) | |||
|
71 | ||||
|
72 | access_log_format = loki_access_log_format | |||
|
73 | if os.environ.get('RC_LOGGING_FORMATTER') == 'json': | |||
|
74 | access_log_format = json_access_log_format | |||
|
75 | ||||
51 | # self adjust workers based on CPU count, to use maximum of CPU and not overquota the resources |
|
76 | # self adjust workers based on CPU count, to use maximum of CPU and not overquota the resources | |
52 | # workers = get_workers() |
|
77 | # workers = get_workers() | |
53 |
|
78 | |||
54 | # Gunicorn access log level |
|
79 | # Gunicorn access log level | |
55 | loglevel = 'info' |
|
80 | loglevel = 'info' | |
56 |
|
81 | |||
57 | # Process name visible in a process list |
|
82 | # Process name visible in a process list | |
58 | proc_name = "rhodecode_vcsserver" |
|
83 | proc_name = "rhodecode_vcsserver" | |
59 |
|
84 | |||
60 | # Type of worker class, one of `sync`, `gevent` or `gthread` |
|
85 | # Type of worker class, one of `sync`, `gevent` or `gthread` | |
61 | # currently `sync` is the only option allowed for vcsserver and for rhodecode all of 3 are allowed |
|
86 | # currently `sync` is the only option allowed for vcsserver and for rhodecode all of 3 are allowed | |
62 | # gevent: |
|
87 | # gevent: | |
63 | # In this case, the maximum number of concurrent requests is (N workers * X worker_connections) |
|
88 | # In this case, the maximum number of concurrent requests is (N workers * X worker_connections) | |
64 | # e.g. workers =3 worker_connections=10 = 3*10, 30 concurrent requests can be handled |
|
89 | # e.g. workers =3 worker_connections=10 = 3*10, 30 concurrent requests can be handled | |
65 | # gthread: |
|
90 | # gthread: | |
66 | # In this case, the maximum number of concurrent requests is (N workers * X threads) |
|
91 | # In this case, the maximum number of concurrent requests is (N workers * X threads) | |
67 | # e.g. workers = 3 threads=3 = 3*3, 9 concurrent requests can be handled |
|
92 | # e.g. workers = 3 threads=3 = 3*3, 9 concurrent requests can be handled | |
68 | worker_class = 'sync' |
|
93 | worker_class = 'sync' | |
69 |
|
94 | |||
70 | # Sets the number of process workers. More workers means more concurrent connections |
|
95 | # Sets the number of process workers. More workers means more concurrent connections | |
71 | # RhodeCode can handle at the same time. Each additional worker also it increases |
|
96 | # RhodeCode can handle at the same time. Each additional worker also it increases | |
72 | # memory usage as each has its own set of caches. |
|
97 | # memory usage as each has its own set of caches. | |
73 | # The Recommended value is (2 * NUMBER_OF_CPUS + 1), eg 2CPU = 5 workers, but no more |
|
98 | # The Recommended value is (2 * NUMBER_OF_CPUS + 1), eg 2CPU = 5 workers, but no more | |
74 | # than 8-10 unless for huge deployments .e.g 700-1000 users. |
|
99 | # than 8-10 unless for huge deployments .e.g 700-1000 users. | |
75 | # `instance_id = *` must be set in the [app:main] section below (which is the default) |
|
100 | # `instance_id = *` must be set in the [app:main] section below (which is the default) | |
76 | # when using more than 1 worker. |
|
101 | # when using more than 1 worker. | |
77 | workers = 2 |
|
102 | workers = 2 | |
78 |
|
103 | |||
79 | # Threads numbers for worker class gthread |
|
104 | # Threads numbers for worker class gthread | |
80 | threads = 1 |
|
105 | threads = 1 | |
81 |
|
106 | |||
82 | # The maximum number of simultaneous clients. Valid only for gevent |
|
107 | # The maximum number of simultaneous clients. Valid only for gevent | |
83 | # In this case, the maximum number of concurrent requests is (N workers * X worker_connections) |
|
108 | # In this case, the maximum number of concurrent requests is (N workers * X worker_connections) | |
84 | # e.g workers =3 worker_connections=10 = 3*10, 30 concurrent requests can be handled |
|
109 | # e.g workers =3 worker_connections=10 = 3*10, 30 concurrent requests can be handled | |
85 | worker_connections = 10 |
|
110 | worker_connections = 10 | |
86 |
|
111 | |||
87 | # Max number of requests that worker will handle before being gracefully restarted. |
|
112 | # Max number of requests that worker will handle before being gracefully restarted. | |
88 | # Prevents memory leaks, jitter adds variability so not all workers are restarted at once. |
|
113 | # Prevents memory leaks, jitter adds variability so not all workers are restarted at once. | |
89 | max_requests = 2000 |
|
114 | max_requests = 2000 | |
90 | max_requests_jitter = int(max_requests * 0.2) # 20% of max_requests |
|
115 | max_requests_jitter = int(max_requests * 0.2) # 20% of max_requests | |
91 |
|
116 | |||
92 | # The maximum number of pending connections. |
|
117 | # The maximum number of pending connections. | |
93 | # Exceeding this number results in the client getting an error when attempting to connect. |
|
118 | # Exceeding this number results in the client getting an error when attempting to connect. | |
94 | backlog = 64 |
|
119 | backlog = 64 | |
95 |
|
120 | |||
96 | # The Amount of time a worker can spend with handling a request before it |
|
121 | # The Amount of time a worker can spend with handling a request before it | |
97 | # gets killed and restarted. By default, set to 21600 (6hrs) |
|
122 | # gets killed and restarted. By default, set to 21600 (6hrs) | |
98 | # Examples: 1800 (30min), 3600 (1hr), 7200 (2hr), 43200 (12h) |
|
123 | # Examples: 1800 (30min), 3600 (1hr), 7200 (2hr), 43200 (12h) | |
99 | timeout = 21600 |
|
124 | timeout = 21600 | |
100 |
|
125 | |||
101 | # The maximum size of HTTP request line in bytes. |
|
126 | # The maximum size of HTTP request line in bytes. | |
102 | # 0 for unlimited |
|
127 | # 0 for unlimited | |
103 | limit_request_line = 0 |
|
128 | limit_request_line = 0 | |
104 |
|
129 | |||
105 | # Limit the number of HTTP headers fields in a request. |
|
130 | # Limit the number of HTTP headers fields in a request. | |
106 | # By default this value is 100 and can't be larger than 32768. |
|
131 | # By default this value is 100 and can't be larger than 32768. | |
107 | limit_request_fields = 32768 |
|
132 | limit_request_fields = 32768 | |
108 |
|
133 | |||
109 | # Limit the allowed size of an HTTP request header field. |
|
134 | # Limit the allowed size of an HTTP request header field. | |
110 | # Value is a positive number or 0. |
|
135 | # Value is a positive number or 0. | |
111 | # Setting it to 0 will allow unlimited header field sizes. |
|
136 | # Setting it to 0 will allow unlimited header field sizes. | |
112 | limit_request_field_size = 0 |
|
137 | limit_request_field_size = 0 | |
113 |
|
138 | |||
114 | # Timeout for graceful workers restart. |
|
139 | # Timeout for graceful workers restart. | |
115 | # After receiving a restart signal, workers have this much time to finish |
|
140 | # After receiving a restart signal, workers have this much time to finish | |
116 | # serving requests. Workers still alive after the timeout (starting from the |
|
141 | # serving requests. Workers still alive after the timeout (starting from the | |
117 | # receipt of the restart signal) are force killed. |
|
142 | # receipt of the restart signal) are force killed. | |
118 | # Examples: 1800 (30min), 3600 (1hr), 7200 (2hr), 43200 (12h) |
|
143 | # Examples: 1800 (30min), 3600 (1hr), 7200 (2hr), 43200 (12h) | |
119 | graceful_timeout = 21600 |
|
144 | graceful_timeout = 21600 | |
120 |
|
145 | |||
121 | # The number of seconds to wait for requests on a Keep-Alive connection. |
|
146 | # The number of seconds to wait for requests on a Keep-Alive connection. | |
122 | # Generally set in the 1-5 seconds range. |
|
147 | # Generally set in the 1-5 seconds range. | |
123 | keepalive = 2 |
|
148 | keepalive = 2 | |
124 |
|
149 | |||
125 | # Maximum memory usage that each worker can use before it will receive a |
|
150 | # Maximum memory usage that each worker can use before it will receive a | |
126 | # graceful restart signal 0 = memory monitoring is disabled |
|
151 | # graceful restart signal 0 = memory monitoring is disabled | |
127 | # Examples: 268435456 (256MB), 536870912 (512MB) |
|
152 | # Examples: 268435456 (256MB), 536870912 (512MB) | |
128 | # 1073741824 (1GB), 2147483648 (2GB), 4294967296 (4GB) |
|
153 | # 1073741824 (1GB), 2147483648 (2GB), 4294967296 (4GB) | |
129 | # Dynamic formula 1024 * 1024 * 256 == 256MBs |
|
154 | # Dynamic formula 1024 * 1024 * 256 == 256MBs | |
130 | memory_max_usage = 0 |
|
155 | memory_max_usage = 0 | |
131 |
|
156 | |||
132 | # How often in seconds to check for memory usage for each gunicorn worker |
|
157 | # How often in seconds to check for memory usage for each gunicorn worker | |
133 | memory_usage_check_interval = 60 |
|
158 | memory_usage_check_interval = 60 | |
134 |
|
159 | |||
135 | # Threshold value for which we don't recycle worker if GarbageCollection |
|
160 | # Threshold value for which we don't recycle worker if GarbageCollection | |
136 | # frees up enough resources. Before each restart, we try to run GC on worker |
|
161 | # frees up enough resources. Before each restart, we try to run GC on worker | |
137 | # in case we get enough free memory after that; restart will not happen. |
|
162 | # in case we get enough free memory after that; restart will not happen. | |
138 | memory_usage_recovery_threshold = 0.8 |
|
163 | memory_usage_recovery_threshold = 0.8 | |
139 |
|
164 | |||
140 |
|
165 | |||
141 | @dataclasses.dataclass |
|
166 | @dataclasses.dataclass | |
142 | class MemoryCheckConfig: |
|
167 | class MemoryCheckConfig: | |
143 | max_usage: int |
|
168 | max_usage: int | |
144 | check_interval: int |
|
169 | check_interval: int | |
145 | recovery_threshold: float |
|
170 | recovery_threshold: float | |
146 |
|
171 | |||
147 |
|
172 | |||
148 | def _get_process_rss(pid=None): |
|
173 | def _get_process_rss(pid=None): | |
149 | try: |
|
174 | try: | |
150 | import psutil |
|
175 | import psutil | |
151 | if pid: |
|
176 | if pid: | |
152 | proc = psutil.Process(pid) |
|
177 | proc = psutil.Process(pid) | |
153 | else: |
|
178 | else: | |
154 | proc = psutil.Process() |
|
179 | proc = psutil.Process() | |
155 | return proc.memory_info().rss |
|
180 | return proc.memory_info().rss | |
156 | except Exception: |
|
181 | except Exception: | |
157 | return None |
|
182 | return None | |
158 |
|
183 | |||
159 |
|
184 | |||
160 | def _get_config(ini_path): |
|
185 | def _get_config(ini_path): | |
161 | import configparser |
|
186 | import configparser | |
162 |
|
187 | |||
163 | try: |
|
188 | try: | |
164 | config = configparser.RawConfigParser() |
|
189 | config = configparser.RawConfigParser() | |
165 | config.read(ini_path) |
|
190 | config.read(ini_path) | |
166 | return config |
|
191 | return config | |
167 | except Exception: |
|
192 | except Exception: | |
168 | return None |
|
193 | return None | |
169 |
|
194 | |||
170 |
|
195 | |||
171 | def get_memory_usage_params(config=None): |
|
196 | def get_memory_usage_params(config=None): | |
172 | # memory spec defaults |
|
197 | # memory spec defaults | |
173 | _memory_max_usage = memory_max_usage |
|
198 | _memory_max_usage = memory_max_usage | |
174 | _memory_usage_check_interval = memory_usage_check_interval |
|
199 | _memory_usage_check_interval = memory_usage_check_interval | |
175 | _memory_usage_recovery_threshold = memory_usage_recovery_threshold |
|
200 | _memory_usage_recovery_threshold = memory_usage_recovery_threshold | |
176 |
|
201 | |||
177 | if config: |
|
202 | if config: | |
178 | ini_path = os.path.abspath(config) |
|
203 | ini_path = os.path.abspath(config) | |
179 | conf = _get_config(ini_path) |
|
204 | conf = _get_config(ini_path) | |
180 |
|
205 | |||
181 | section = 'server:main' |
|
206 | section = 'server:main' | |
182 | if conf and conf.has_section(section): |
|
207 | if conf and conf.has_section(section): | |
183 |
|
208 | |||
184 | if conf.has_option(section, 'memory_max_usage'): |
|
209 | if conf.has_option(section, 'memory_max_usage'): | |
185 | _memory_max_usage = conf.getint(section, 'memory_max_usage') |
|
210 | _memory_max_usage = conf.getint(section, 'memory_max_usage') | |
186 |
|
211 | |||
187 | if conf.has_option(section, 'memory_usage_check_interval'): |
|
212 | if conf.has_option(section, 'memory_usage_check_interval'): | |
188 | _memory_usage_check_interval = conf.getint(section, 'memory_usage_check_interval') |
|
213 | _memory_usage_check_interval = conf.getint(section, 'memory_usage_check_interval') | |
189 |
|
214 | |||
190 | if conf.has_option(section, 'memory_usage_recovery_threshold'): |
|
215 | if conf.has_option(section, 'memory_usage_recovery_threshold'): | |
191 | _memory_usage_recovery_threshold = conf.getfloat(section, 'memory_usage_recovery_threshold') |
|
216 | _memory_usage_recovery_threshold = conf.getfloat(section, 'memory_usage_recovery_threshold') | |
192 |
|
217 | |||
193 | _memory_max_usage = int(os.environ.get('RC_GUNICORN_MEMORY_MAX_USAGE', '') |
|
218 | _memory_max_usage = int(os.environ.get('RC_GUNICORN_MEMORY_MAX_USAGE', '') | |
194 | or _memory_max_usage) |
|
219 | or _memory_max_usage) | |
195 | _memory_usage_check_interval = int(os.environ.get('RC_GUNICORN_MEMORY_USAGE_CHECK_INTERVAL', '') |
|
220 | _memory_usage_check_interval = int(os.environ.get('RC_GUNICORN_MEMORY_USAGE_CHECK_INTERVAL', '') | |
196 | or _memory_usage_check_interval) |
|
221 | or _memory_usage_check_interval) | |
197 | _memory_usage_recovery_threshold = float(os.environ.get('RC_GUNICORN_MEMORY_USAGE_RECOVERY_THRESHOLD', '') |
|
222 | _memory_usage_recovery_threshold = float(os.environ.get('RC_GUNICORN_MEMORY_USAGE_RECOVERY_THRESHOLD', '') | |
198 | or _memory_usage_recovery_threshold) |
|
223 | or _memory_usage_recovery_threshold) | |
199 |
|
224 | |||
200 | return MemoryCheckConfig(_memory_max_usage, _memory_usage_check_interval, _memory_usage_recovery_threshold) |
|
225 | return MemoryCheckConfig(_memory_max_usage, _memory_usage_check_interval, _memory_usage_recovery_threshold) | |
201 |
|
226 | |||
202 |
|
227 | |||
203 | def _time_with_offset(check_interval): |
|
228 | def _time_with_offset(check_interval): | |
204 | return time.time() - random.randint(0, check_interval/2.0) |
|
229 | return time.time() - random.randint(0, check_interval/2.0) | |
205 |
|
230 | |||
206 |
|
231 | |||
207 | def pre_fork(server, worker): |
|
232 | def pre_fork(server, worker): | |
208 | pass |
|
233 | pass | |
209 |
|
234 | |||
210 |
|
235 | |||
211 | def post_fork(server, worker): |
|
236 | def post_fork(server, worker): | |
212 |
|
237 | |||
213 | memory_conf = get_memory_usage_params() |
|
238 | memory_conf = get_memory_usage_params() | |
214 | _memory_max_usage = memory_conf.max_usage |
|
239 | _memory_max_usage = memory_conf.max_usage | |
215 | _memory_usage_check_interval = memory_conf.check_interval |
|
240 | _memory_usage_check_interval = memory_conf.check_interval | |
216 | _memory_usage_recovery_threshold = memory_conf.recovery_threshold |
|
241 | _memory_usage_recovery_threshold = memory_conf.recovery_threshold | |
217 |
|
242 | |||
218 | worker._memory_max_usage = int(os.environ.get('RC_GUNICORN_MEMORY_MAX_USAGE', '') |
|
243 | worker._memory_max_usage = int(os.environ.get('RC_GUNICORN_MEMORY_MAX_USAGE', '') | |
219 | or _memory_max_usage) |
|
244 | or _memory_max_usage) | |
220 | worker._memory_usage_check_interval = int(os.environ.get('RC_GUNICORN_MEMORY_USAGE_CHECK_INTERVAL', '') |
|
245 | worker._memory_usage_check_interval = int(os.environ.get('RC_GUNICORN_MEMORY_USAGE_CHECK_INTERVAL', '') | |
221 | or _memory_usage_check_interval) |
|
246 | or _memory_usage_check_interval) | |
222 | worker._memory_usage_recovery_threshold = float(os.environ.get('RC_GUNICORN_MEMORY_USAGE_RECOVERY_THRESHOLD', '') |
|
247 | worker._memory_usage_recovery_threshold = float(os.environ.get('RC_GUNICORN_MEMORY_USAGE_RECOVERY_THRESHOLD', '') | |
223 | or _memory_usage_recovery_threshold) |
|
248 | or _memory_usage_recovery_threshold) | |
224 |
|
249 | |||
225 | # register memory last check time, with some random offset so we don't recycle all |
|
250 | # register memory last check time, with some random offset so we don't recycle all | |
226 | # at once |
|
251 | # at once | |
227 | worker._last_memory_check_time = _time_with_offset(_memory_usage_check_interval) |
|
252 | worker._last_memory_check_time = _time_with_offset(_memory_usage_check_interval) | |
228 |
|
253 | |||
229 | if _memory_max_usage: |
|
254 | if _memory_max_usage: | |
230 | server.log.info("pid=[%-10s] WORKER spawned with max memory set at %s", worker.pid, |
|
255 | server.log.info("pid=[%-10s] WORKER spawned with max memory set at %s", worker.pid, | |
231 | _format_data_size(_memory_max_usage)) |
|
256 | _format_data_size(_memory_max_usage)) | |
232 | else: |
|
257 | else: | |
233 | server.log.info("pid=[%-10s] WORKER spawned", worker.pid) |
|
258 | server.log.info("pid=[%-10s] WORKER spawned", worker.pid) | |
234 |
|
259 | |||
235 |
|
260 | |||
236 | def pre_exec(server): |
|
261 | def pre_exec(server): | |
237 | server.log.info("Forked child, re-executing.") |
|
262 | server.log.info("Forked child, re-executing.") | |
238 |
|
263 | |||
239 |
|
264 | |||
240 | def on_starting(server): |
|
265 | def on_starting(server): | |
241 | server_lbl = '{} {}'.format(server.proc_name, server.address) |
|
266 | server_lbl = '{} {}'.format(server.proc_name, server.address) | |
242 | server.log.info("Server %s is starting.", server_lbl) |
|
267 | server.log.info("Server %s is starting.", server_lbl) | |
243 | server.log.info('Config:') |
|
268 | server.log.info('Config:') | |
244 | server.log.info(f"\n{server.cfg}") |
|
269 | server.log.info(f"\n{server.cfg}") | |
245 | server.log.info(get_memory_usage_params()) |
|
270 | server.log.info(get_memory_usage_params()) | |
246 |
|
271 | |||
247 |
|
272 | |||
248 | def when_ready(server): |
|
273 | def when_ready(server): | |
249 | server.log.info("Server %s is ready. Spawning workers", server) |
|
274 | server.log.info("Server %s is ready. Spawning workers", server) | |
250 |
|
275 | |||
251 |
|
276 | |||
252 | def on_reload(server): |
|
277 | def on_reload(server): | |
253 | pass |
|
278 | pass | |
254 |
|
279 | |||
255 |
|
280 | |||
256 | def _format_data_size(size, unit="B", precision=1, binary=True): |
|
281 | def _format_data_size(size, unit="B", precision=1, binary=True): | |
257 | """Format a number using SI units (kilo, mega, etc.). |
|
282 | """Format a number using SI units (kilo, mega, etc.). | |
258 |
|
283 | |||
259 | ``size``: The number as a float or int. |
|
284 | ``size``: The number as a float or int. | |
260 |
|
285 | |||
261 | ``unit``: The unit name in plural form. Examples: "bytes", "B". |
|
286 | ``unit``: The unit name in plural form. Examples: "bytes", "B". | |
262 |
|
287 | |||
263 | ``precision``: How many digits to the right of the decimal point. Default |
|
288 | ``precision``: How many digits to the right of the decimal point. Default | |
264 | is 1. 0 suppresses the decimal point. |
|
289 | is 1. 0 suppresses the decimal point. | |
265 |
|
290 | |||
266 | ``binary``: If false, use base-10 decimal prefixes (kilo = K = 1000). |
|
291 | ``binary``: If false, use base-10 decimal prefixes (kilo = K = 1000). | |
267 | If true, use base-2 binary prefixes (kibi = Ki = 1024). |
|
292 | If true, use base-2 binary prefixes (kibi = Ki = 1024). | |
268 |
|
293 | |||
269 | ``full_name``: If false (default), use the prefix abbreviation ("k" or |
|
294 | ``full_name``: If false (default), use the prefix abbreviation ("k" or | |
270 | "Ki"). If true, use the full prefix ("kilo" or "kibi"). If false, |
|
295 | "Ki"). If true, use the full prefix ("kilo" or "kibi"). If false, | |
271 | use abbreviation ("k" or "Ki"). |
|
296 | use abbreviation ("k" or "Ki"). | |
272 |
|
297 | |||
273 | """ |
|
298 | """ | |
274 |
|
299 | |||
275 | if not binary: |
|
300 | if not binary: | |
276 | base = 1000 |
|
301 | base = 1000 | |
277 | multiples = ('', 'k', 'M', 'G', 'T', 'P', 'E', 'Z', 'Y') |
|
302 | multiples = ('', 'k', 'M', 'G', 'T', 'P', 'E', 'Z', 'Y') | |
278 | else: |
|
303 | else: | |
279 | base = 1024 |
|
304 | base = 1024 | |
280 | multiples = ('', 'Ki', 'Mi', 'Gi', 'Ti', 'Pi', 'Ei', 'Zi', 'Yi') |
|
305 | multiples = ('', 'Ki', 'Mi', 'Gi', 'Ti', 'Pi', 'Ei', 'Zi', 'Yi') | |
281 |
|
306 | |||
282 | sign = "" |
|
307 | sign = "" | |
283 | if size > 0: |
|
308 | if size > 0: | |
284 | m = int(math.log(size, base)) |
|
309 | m = int(math.log(size, base)) | |
285 | elif size < 0: |
|
310 | elif size < 0: | |
286 | sign = "-" |
|
311 | sign = "-" | |
287 | size = -size |
|
312 | size = -size | |
288 | m = int(math.log(size, base)) |
|
313 | m = int(math.log(size, base)) | |
289 | else: |
|
314 | else: | |
290 | m = 0 |
|
315 | m = 0 | |
291 | if m > 8: |
|
316 | if m > 8: | |
292 | m = 8 |
|
317 | m = 8 | |
293 |
|
318 | |||
294 | if m == 0: |
|
319 | if m == 0: | |
295 | precision = '%.0f' |
|
320 | precision = '%.0f' | |
296 | else: |
|
321 | else: | |
297 | precision = '%%.%df' % precision |
|
322 | precision = '%%.%df' % precision | |
298 |
|
323 | |||
299 | size = precision % (size / math.pow(base, m)) |
|
324 | size = precision % (size / math.pow(base, m)) | |
300 |
|
325 | |||
301 | return '%s%s %s%s' % (sign, size.strip(), multiples[m], unit) |
|
326 | return '%s%s %s%s' % (sign, size.strip(), multiples[m], unit) | |
302 |
|
327 | |||
303 |
|
328 | |||
304 | def _check_memory_usage(worker): |
|
329 | def _check_memory_usage(worker): | |
305 | _memory_max_usage = worker._memory_max_usage |
|
330 | _memory_max_usage = worker._memory_max_usage | |
306 | if not _memory_max_usage: |
|
331 | if not _memory_max_usage: | |
307 | return |
|
332 | return | |
308 |
|
333 | |||
309 | _memory_usage_check_interval = worker._memory_usage_check_interval |
|
334 | _memory_usage_check_interval = worker._memory_usage_check_interval | |
310 | _memory_usage_recovery_threshold = memory_max_usage * worker._memory_usage_recovery_threshold |
|
335 | _memory_usage_recovery_threshold = memory_max_usage * worker._memory_usage_recovery_threshold | |
311 |
|
336 | |||
312 | elapsed = time.time() - worker._last_memory_check_time |
|
337 | elapsed = time.time() - worker._last_memory_check_time | |
313 | if elapsed > _memory_usage_check_interval: |
|
338 | if elapsed > _memory_usage_check_interval: | |
314 | mem_usage = _get_process_rss() |
|
339 | mem_usage = _get_process_rss() | |
315 | if mem_usage and mem_usage > _memory_max_usage: |
|
340 | if mem_usage and mem_usage > _memory_max_usage: | |
316 | worker.log.info( |
|
341 | worker.log.info( | |
317 | "memory usage %s > %s, forcing gc", |
|
342 | "memory usage %s > %s, forcing gc", | |
318 | _format_data_size(mem_usage), _format_data_size(_memory_max_usage)) |
|
343 | _format_data_size(mem_usage), _format_data_size(_memory_max_usage)) | |
319 | # Try to clean it up by forcing a full collection. |
|
344 | # Try to clean it up by forcing a full collection. | |
320 | gc.collect() |
|
345 | gc.collect() | |
321 | mem_usage = _get_process_rss() |
|
346 | mem_usage = _get_process_rss() | |
322 | if mem_usage > _memory_usage_recovery_threshold: |
|
347 | if mem_usage > _memory_usage_recovery_threshold: | |
323 | # Didn't clean up enough, we'll have to terminate. |
|
348 | # Didn't clean up enough, we'll have to terminate. | |
324 | worker.log.warning( |
|
349 | worker.log.warning( | |
325 | "memory usage %s > %s after gc, quitting", |
|
350 | "memory usage %s > %s after gc, quitting", | |
326 | _format_data_size(mem_usage), _format_data_size(_memory_max_usage)) |
|
351 | _format_data_size(mem_usage), _format_data_size(_memory_max_usage)) | |
327 | # This will cause worker to auto-restart itself |
|
352 | # This will cause worker to auto-restart itself | |
328 | worker.alive = False |
|
353 | worker.alive = False | |
329 | worker._last_memory_check_time = time.time() |
|
354 | worker._last_memory_check_time = time.time() | |
330 |
|
355 | |||
331 |
|
356 | |||
332 | def worker_int(worker): |
|
357 | def worker_int(worker): | |
333 | worker.log.info("pid=[%-10s] worker received INT or QUIT signal", worker.pid) |
|
358 | worker.log.info("pid=[%-10s] worker received INT or QUIT signal", worker.pid) | |
334 |
|
359 | |||
335 | # get traceback info, when a worker crashes |
|
360 | # get traceback info, when a worker crashes | |
336 | def get_thread_id(t_id): |
|
361 | def get_thread_id(t_id): | |
337 | id2name = dict([(th.ident, th.name) for th in threading.enumerate()]) |
|
362 | id2name = dict([(th.ident, th.name) for th in threading.enumerate()]) | |
338 | return id2name.get(t_id, "unknown_thread_id") |
|
363 | return id2name.get(t_id, "unknown_thread_id") | |
339 |
|
364 | |||
340 | code = [] |
|
365 | code = [] | |
341 | for thread_id, stack in sys._current_frames().items(): # noqa |
|
366 | for thread_id, stack in sys._current_frames().items(): # noqa | |
342 | code.append( |
|
367 | code.append( | |
343 | "\n# Thread: %s(%d)" % (get_thread_id(thread_id), thread_id)) |
|
368 | "\n# Thread: %s(%d)" % (get_thread_id(thread_id), thread_id)) | |
344 | for fname, lineno, name, line in traceback.extract_stack(stack): |
|
369 | for fname, lineno, name, line in traceback.extract_stack(stack): | |
345 | code.append('File: "%s", line %d, in %s' % (fname, lineno, name)) |
|
370 | code.append('File: "%s", line %d, in %s' % (fname, lineno, name)) | |
346 | if line: |
|
371 | if line: | |
347 | code.append(" %s" % (line.strip())) |
|
372 | code.append(" %s" % (line.strip())) | |
348 | worker.log.debug("\n".join(code)) |
|
373 | worker.log.debug("\n".join(code)) | |
349 |
|
374 | |||
350 |
|
375 | |||
351 | def worker_abort(worker): |
|
376 | def worker_abort(worker): | |
352 | worker.log.info("pid=[%-10s] worker received SIGABRT signal", worker.pid) |
|
377 | worker.log.info("pid=[%-10s] worker received SIGABRT signal", worker.pid) | |
353 |
|
378 | |||
354 |
|
379 | |||
355 | def worker_exit(server, worker): |
|
380 | def worker_exit(server, worker): | |
356 | worker.log.info("pid=[%-10s] worker exit", worker.pid) |
|
381 | worker.log.info("pid=[%-10s] worker exit", worker.pid) | |
357 |
|
382 | |||
358 |
|
383 | |||
359 | def child_exit(server, worker): |
|
384 | def child_exit(server, worker): | |
360 | worker.log.info("pid=[%-10s] worker child exit", worker.pid) |
|
385 | worker.log.info("pid=[%-10s] worker child exit", worker.pid) | |
361 |
|
386 | |||
362 |
|
387 | |||
363 | def pre_request(worker, req): |
|
388 | def pre_request(worker, req): | |
364 | worker.start_time = time.time() |
|
389 | worker.start_time = time.time() | |
365 | worker.log.debug( |
|
390 | worker.log.debug( | |
366 | "GNCRN PRE WORKER [cnt:%s]: %s %s", worker.nr, req.method, req.path) |
|
391 | "GNCRN PRE WORKER [cnt:%s]: %s %s", worker.nr, req.method, req.path) | |
367 |
|
392 | |||
368 |
|
393 | |||
369 | def post_request(worker, req, environ, resp): |
|
394 | def post_request(worker, req, environ, resp): | |
370 | total_time = time.time() - worker.start_time |
|
395 | total_time = time.time() - worker.start_time | |
371 | # Gunicorn sometimes has problems with reading the status_code |
|
396 | # Gunicorn sometimes has problems with reading the status_code | |
372 | status_code = getattr(resp, 'status_code', '') |
|
397 | status_code = getattr(resp, 'status_code', '') | |
373 | worker.log.debug( |
|
398 | worker.log.debug( | |
374 | "GNCRN POST WORKER [cnt:%s]: %s %s resp: %s, Load Time: %.4fs", |
|
399 | "GNCRN POST WORKER [cnt:%s]: %s %s resp: %s, Load Time: %.4fs", | |
375 | worker.nr, req.method, req.path, status_code, total_time) |
|
400 | worker.nr, req.method, req.path, status_code, total_time) | |
376 | _check_memory_usage(worker) |
|
401 | _check_memory_usage(worker) | |
377 |
|
402 | |||
378 |
|
403 | |||
379 | def _filter_proxy(ip): |
|
404 | def _filter_proxy(ip): | |
380 | """ |
|
405 | """ | |
381 | Passed in IP addresses in HEADERS can be in a special format of multiple |
|
406 | Passed in IP addresses in HEADERS can be in a special format of multiple | |
382 | ips. Those comma separated IPs are passed from various proxies in the |
|
407 | ips. Those comma separated IPs are passed from various proxies in the | |
383 | chain of request processing. The left-most being the original client. |
|
408 | chain of request processing. The left-most being the original client. | |
384 | We only care about the first IP which came from the org. client. |
|
409 | We only care about the first IP which came from the org. client. | |
385 |
|
410 | |||
386 | :param ip: ip string from headers |
|
411 | :param ip: ip string from headers | |
387 | """ |
|
412 | """ | |
388 | if ',' in ip: |
|
413 | if ',' in ip: | |
389 | _ips = ip.split(',') |
|
414 | _ips = ip.split(',') | |
390 | _first_ip = _ips[0].strip() |
|
415 | _first_ip = _ips[0].strip() | |
391 | return _first_ip |
|
416 | return _first_ip | |
392 | return ip |
|
417 | return ip | |
393 |
|
418 | |||
394 |
|
419 | |||
395 | def _filter_port(ip): |
|
420 | def _filter_port(ip): | |
396 | """ |
|
421 | """ | |
397 | Removes a port from ip, there are 4 main cases to handle here. |
|
422 | Removes a port from ip, there are 4 main cases to handle here. | |
398 | - ipv4 eg. 127.0.0.1 |
|
423 | - ipv4 eg. 127.0.0.1 | |
399 | - ipv6 eg. ::1 |
|
424 | - ipv6 eg. ::1 | |
400 | - ipv4+port eg. 127.0.0.1:8080 |
|
425 | - ipv4+port eg. 127.0.0.1:8080 | |
401 | - ipv6+port eg. [::1]:8080 |
|
426 | - ipv6+port eg. [::1]:8080 | |
402 |
|
427 | |||
403 | :param ip: |
|
428 | :param ip: | |
404 | """ |
|
429 | """ | |
405 | def is_ipv6(ip_addr): |
|
430 | def is_ipv6(ip_addr): | |
406 | if hasattr(socket, 'inet_pton'): |
|
431 | if hasattr(socket, 'inet_pton'): | |
407 | try: |
|
432 | try: | |
408 | socket.inet_pton(socket.AF_INET6, ip_addr) |
|
433 | socket.inet_pton(socket.AF_INET6, ip_addr) | |
409 | except socket.error: |
|
434 | except socket.error: | |
410 | return False |
|
435 | return False | |
411 | else: |
|
436 | else: | |
412 | return False |
|
437 | return False | |
413 | return True |
|
438 | return True | |
414 |
|
439 | |||
415 | if ':' not in ip: # must be ipv4 pure ip |
|
440 | if ':' not in ip: # must be ipv4 pure ip | |
416 | return ip |
|
441 | return ip | |
417 |
|
442 | |||
418 | if '[' in ip and ']' in ip: # ipv6 with port |
|
443 | if '[' in ip and ']' in ip: # ipv6 with port | |
419 | return ip.split(']')[0][1:].lower() |
|
444 | return ip.split(']')[0][1:].lower() | |
420 |
|
445 | |||
421 | # must be ipv6 or ipv4 with port |
|
446 | # must be ipv6 or ipv4 with port | |
422 | if is_ipv6(ip): |
|
447 | if is_ipv6(ip): | |
423 | return ip |
|
448 | return ip | |
424 | else: |
|
449 | else: | |
425 | ip, _port = ip.split(':')[:2] # means ipv4+port |
|
450 | ip, _port = ip.split(':')[:2] # means ipv4+port | |
426 | return ip |
|
451 | return ip | |
427 |
|
452 | |||
428 |
|
453 | |||
429 | def get_ip_addr(environ): |
|
454 | def get_ip_addr(environ): | |
430 | proxy_key = 'HTTP_X_REAL_IP' |
|
455 | proxy_key = 'HTTP_X_REAL_IP' | |
431 | proxy_key2 = 'HTTP_X_FORWARDED_FOR' |
|
456 | proxy_key2 = 'HTTP_X_FORWARDED_FOR' | |
432 | def_key = 'REMOTE_ADDR' |
|
457 | def_key = 'REMOTE_ADDR' | |
433 |
|
458 | |||
434 | def _filters(x): |
|
459 | def _filters(x): | |
435 | return _filter_port(_filter_proxy(x)) |
|
460 | return _filter_port(_filter_proxy(x)) | |
436 |
|
461 | |||
437 | ip = environ.get(proxy_key) |
|
462 | ip = environ.get(proxy_key) | |
438 | if ip: |
|
463 | if ip: | |
439 | return _filters(ip) |
|
464 | return _filters(ip) | |
440 |
|
465 | |||
441 | ip = environ.get(proxy_key2) |
|
466 | ip = environ.get(proxy_key2) | |
442 | if ip: |
|
467 | if ip: | |
443 | return _filters(ip) |
|
468 | return _filters(ip) | |
444 |
|
469 | |||
445 | ip = environ.get(def_key, '0.0.0.0') |
|
470 | ip = environ.get(def_key, '0.0.0.0') | |
446 | return _filters(ip) |
|
471 | return _filters(ip) | |
447 |
|
472 | |||
448 |
|
473 | |||
449 | class RhodeCodeLogger(Logger): |
|
474 | class RhodeCodeLogger(Logger): | |
450 | """ |
|
475 | """ | |
451 | Custom Logger that allows some customization that gunicorn doesn't allow |
|
476 | Custom Logger that allows some customization that gunicorn doesn't allow | |
452 | """ |
|
477 | """ | |
453 |
|
478 | |||
454 | datefmt = r"%Y-%m-%d %H:%M:%S" |
|
479 | datefmt = r"%Y-%m-%d %H:%M:%S" | |
455 |
|
480 | |||
456 | def __init__(self, cfg): |
|
481 | def __init__(self, cfg): | |
457 | Logger.__init__(self, cfg) |
|
482 | Logger.__init__(self, cfg) | |
458 |
|
483 | |||
459 | def now(self): |
|
484 | def now(self): | |
460 | """ return date in RhodeCode Log format """ |
|
485 | """ return date in RhodeCode Log format """ | |
461 | now = time.time() |
|
486 | now = time.time() | |
462 | msecs = int((now - int(now)) * 1000) |
|
487 | msecs = int((now - int(now)) * 1000) | |
463 | return time.strftime(self.datefmt, time.localtime(now)) + '.{0:03d}'.format(msecs) |
|
488 | return time.strftime(self.datefmt, time.localtime(now)) + '.{0:03d}'.format(msecs) | |
464 |
|
489 | |||
465 | def atoms(self, resp, req, environ, request_time): |
|
490 | def atoms(self, resp, req, environ, request_time): | |
466 | """ Gets atoms for log formatting. |
|
491 | """ Gets atoms for log formatting. | |
467 | """ |
|
492 | """ | |
468 | status = resp.status |
|
493 | status = resp.status | |
469 | if isinstance(status, str): |
|
494 | if isinstance(status, str): | |
470 | status = status.split(None, 1)[0] |
|
495 | status = status.split(None, 1)[0] | |
471 | atoms = { |
|
496 | atoms = { | |
472 | 'h': get_ip_addr(environ), |
|
497 | 'h': get_ip_addr(environ), | |
473 | 'l': '-', |
|
498 | 'l': '-', | |
474 | 'u': self._get_user(environ) or '-', |
|
499 | 'u': self._get_user(environ) or '-', | |
475 | 't': self.now(), |
|
500 | 't': self.now(), | |
476 | 'r': "%s %s %s" % (environ['REQUEST_METHOD'], |
|
501 | 'r': "%s %s %s" % (environ['REQUEST_METHOD'], | |
477 | environ['RAW_URI'], |
|
502 | environ['RAW_URI'], | |
478 | environ["SERVER_PROTOCOL"]), |
|
503 | environ["SERVER_PROTOCOL"]), | |
479 | 's': status, |
|
504 | 's': status, | |
480 | 'm': environ.get('REQUEST_METHOD'), |
|
505 | 'm': environ.get('REQUEST_METHOD'), | |
481 | 'U': environ.get('PATH_INFO'), |
|
506 | 'U': environ.get('PATH_INFO'), | |
482 | 'q': environ.get('QUERY_STRING'), |
|
507 | 'q': environ.get('QUERY_STRING'), | |
483 | 'H': environ.get('SERVER_PROTOCOL'), |
|
508 | 'H': environ.get('SERVER_PROTOCOL'), | |
484 | 'b': getattr(resp, 'sent', None) is not None and str(resp.sent) or '-', |
|
509 | 'b': getattr(resp, 'sent', None) is not None and str(resp.sent) or '-', | |
485 | 'B': getattr(resp, 'sent', None), |
|
510 | 'B': getattr(resp, 'sent', None), | |
486 | 'f': environ.get('HTTP_REFERER', '-'), |
|
511 | 'f': environ.get('HTTP_REFERER', '-'), | |
487 | 'a': environ.get('HTTP_USER_AGENT', '-'), |
|
512 | 'a': environ.get('HTTP_USER_AGENT', '-'), | |
488 | 'T': request_time.seconds, |
|
513 | 'T': request_time.seconds, | |
489 | 'D': (request_time.seconds * 1000000) + request_time.microseconds, |
|
514 | 'D': (request_time.seconds * 1000000) + request_time.microseconds, | |
490 | 'M': (request_time.seconds * 1000) + int(request_time.microseconds/1000), |
|
515 | 'M': (request_time.seconds * 1000) + int(request_time.microseconds/1000), | |
491 | 'L': "%d.%06d" % (request_time.seconds, request_time.microseconds), |
|
516 | 'L': "%d.%06d" % (request_time.seconds, request_time.microseconds), | |
492 | 'p': "<%s>" % os.getpid() |
|
517 | 'p': "<%s>" % os.getpid() | |
493 | } |
|
518 | } | |
494 |
|
519 | |||
495 | # add request headers |
|
520 | # add request headers | |
496 | if hasattr(req, 'headers'): |
|
521 | if hasattr(req, 'headers'): | |
497 | req_headers = req.headers |
|
522 | req_headers = req.headers | |
498 | else: |
|
523 | else: | |
499 | req_headers = req |
|
524 | req_headers = req | |
500 |
|
525 | |||
501 | if hasattr(req_headers, "items"): |
|
526 | if hasattr(req_headers, "items"): | |
502 | req_headers = req_headers.items() |
|
527 | req_headers = req_headers.items() | |
503 |
|
528 | |||
504 | atoms.update({"{%s}i" % k.lower(): v for k, v in req_headers}) |
|
529 | atoms.update({"{%s}i" % k.lower(): v for k, v in req_headers}) | |
505 |
|
530 | |||
506 | resp_headers = resp.headers |
|
531 | resp_headers = resp.headers | |
507 | if hasattr(resp_headers, "items"): |
|
532 | if hasattr(resp_headers, "items"): | |
508 | resp_headers = resp_headers.items() |
|
533 | resp_headers = resp_headers.items() | |
509 |
|
534 | |||
510 | # add response headers |
|
535 | # add response headers | |
511 | atoms.update({"{%s}o" % k.lower(): v for k, v in resp_headers}) |
|
536 | atoms.update({"{%s}o" % k.lower(): v for k, v in resp_headers}) | |
512 |
|
537 | |||
513 | # add environ variables |
|
538 | # add environ variables | |
514 | environ_variables = environ.items() |
|
539 | environ_variables = environ.items() | |
515 | atoms.update({"{%s}e" % k.lower(): v for k, v in environ_variables}) |
|
540 | atoms.update({"{%s}e" % k.lower(): v for k, v in environ_variables}) | |
516 |
|
541 | |||
517 | return atoms |
|
542 | return atoms | |
518 |
|
543 | |||
519 |
|
544 | |||
520 | logger_class = RhodeCodeLogger |
|
545 | logger_class = RhodeCodeLogger |
General Comments 0
You need to be logged in to leave comments.
Login now