0FieJddlmZddlZddlZddlZddlZddlZddlZddlm Z m Z ddl m Z m Z ddlmZddlmZddlmZddlmZ ddlZn#e$r ed wxYwejeZd$dZd%dZej d&d'd"Zej d(d)d#ZdS)*) annotationsN)IteratorSequence)AnyLiteral)quote) partition) get_client)Workerz.You have to install memray to use this module. dask_workerr filenamestrkwargsrreturnboolc t|dr|jtj|j|t |jzz }|r| tj |_|j tj|fi|dS)z$Start the memray Tracker on a Server_memrayT)hasattrrclosepathlibPathlocal_directoryridexistsrmdir contextlib ExitStack enter_contextmemrayTracker)r r rpaths >lib/python3.11/site-packages/distributed/diagnostics/memray.py _start_memrayr#s{I&&$!!### < 3 4 43{~CVCV8V WD {{}} $.00K%%t&&v&& 4 report_argsSequence[str] | Literal[False]bytesct|dsdStj|j|t |jzz }|j|`|sX % %j4 #4#44 { " " "$S " " " K= {##tSYYc/>R>R&SSKN; ot $ $wwyys$1BBBE""E&)E&memray-profiles flamegraphz --temporalz--leaksT directorystr | pathlib.Pathworkersint | None | list[str]fetch_reports_parallel bool | int memray_kwargsIterator[None]c +Ktj|}t}|}|d}d|dD}|rt |t r=t|} t |t r|} t|d| }t|}tj j } t|j tfd| i|sJt!jddV|dt}|durt|} n |durd } n|} t'| |D]} | t(| || } | D]\}}|t+t-||d z }|rd }nd}t/t-||zd5}||dddn #1swxYwY#t2$rt4d| YwxYwdS)a(Generate a Memray profile on the workers and download the generated report. Example:: with memray_workers(): client.submit(my_function).result() # Or even while the computation is already running fut = client.submit(my_function) with memray_workers(): time.sleep(10) fut.result() Parameters ---------- directory : str The directory to save the reports to. workers : int | None | list[str] The workers to profile. If int, the first n workers will be used. If None, all workers will be used. If list[str], the workers with the given addresses will be used. report_args : tuple[str] Particularly for native_traces=True, the reports have to be generated on the same host using the same Python interpreter as the profile was generated. Otherwise, native traces will yield unusable results. Therefore, we're generating the reports on the workers and download them afterwards. You can modify the report generation by providing additional arguments and we will generate the reports as:: memray *report_args -f -o .html If the raw data should be fetched instead of the report, set this to False. fetch_reports_parallel : bool | int Fetching results is sometimes slow and it's sometimes not desired to wait for all workers to finish before receiving the first reports. This controls how many workers are fetched concurrently. int: Number of workers to fetch concurrently True: All workers concurrently False: One worker at a time **memray_kwargs Keyword arguments to be passed to memray.Tracker, e.g. {"native_traces": True} r9c&i|]\}}||dS)name).0addrwinfos r" z"memray_workers..s/ +eeFmr$Nr 皙?Texist_okF)r r%r9)safer*.memraywbz2Exception during report downloading from worker %s)rrr scheduler_infoitems isinstanceintlenr.uuiduuid4hexallr0r#valuestimesleepmkdirr r3rrr+write Exceptionlogger exception)r7r9r%r;r=clientrO worker_addr worker_namesnworkersr fetch_parallelwprofilesprofiler!suffixr1s r"memray_workersriFsv Y''I \\F**,,N +K/=i/H/N/N/P/PL /j#../{## gs # # H{##IXI.7mmGz||H zvz-LL(LmLLSSUU V VVVV JsOOO EEE OOTO""" \\F%%W 5 ( (/ ~w / /VV Vzz%!' "H )1(8(8 & &$ W 5\+-F)G)Gb#Q#Q#QQ'$FF&F#d))f,d33&rHHW%%%&&&&&&&&&&&&&&&  & V V V   QST U U U U U V#VVs7BH6H) H6)H- -H60H- 1H66%IIc+Ktj|}t}tjj}|jtfd|i|sJtj ddV| dt}|t||}|dz }|rd}nd }tt||zd 5}||ddddS#1swxYwYdS) aGenerate a Memray profile on the Scheduler and download the generated report. Example:: with memray_scheduler(): client.submit(my_function).result() # Or even while the computation is already running fut = client.submit(my_function) with memray_scheduler(): time.sleep(10) fut.result() Parameters ---------- directory : str The directory to save the reports to. report_args : tuple[str] Particularly for native_traces=True, the reports have to be generated on the same host using the same Python interpreter as the profile was generated. Otherwise, native traces will yield unusable results. Therefore, we're generating the reports on the Scheduler and download them afterwards. You can modify the report generation by providing additional arguments and we will generate the reports as:: memray *report_args -f -o .html If the raw data should be fetched instead of the report, set this to False. **memray_kwargs Keyword arguments to be passed to memray.Tracker, e.g. {"native_traces": True} r rGNTrH)r r% schedulerr*rMrN)rrr rTrUrVrun_on_schedulerr#rYrZr[r3r+rr\) r7r%r=r`r rgr!rhr1s r"memray_schedulerrms[V Y''I \\Fz||H "6 "= U U8 U} U UUUU JsOOO EEE OOTO""" \\F%%&G { "D c$ii& $ ' '2 sC<<DD)r r r rrrrr)r r r rr%r&rr')r4Nr5T) r7r8r9r:r%r&r;r<r=rrr>)r4r5)r7r8r%r&r=rrr>) __future__rrloggingrr/rYrTcollections.abcrrtypingrr urllib.parsertoolz.itertoolzr distributedr distributed.workerr r ImportError getLogger__name__r^r#r3contextmanagerrirmrBr$r"rzs"""""" ........%%%%%%""""""%%%%%%HMMMMHHH +F G GGH  8 $ $"6 $5&*>)- iViViViViVX $5>AAAAAAAs A A